Privacy Policy — AIsuru MCP Servers

AIsuru Agent Link

Source: internal Memori gateway service · Type: exposes the agent as a standard MCP server

Exposes an AIsuru agent as a standard MCP server reachable by external applications (Claude, Cursor, other agents). Connected clients can converse with the agent and search its knowledge; the owner can also manage it remotely (builder tools).

Data retained by the gateway

• Public endpoint identifier and issued access tokens (stored hashed: the clear value is shown only once at creation).
• Technical references to the exposed agent (ID, tenant) and the optional private-agent password, encrypted at-rest.
• Technical call metadata (invoked tool, outcome, timestamp) for operation and diagnostics.

Data NOT retained

• Conversation contents are not duplicated by the gateway: they live in the AIsuru platform under Memori's general privacy policy.

Revocation

The owner can revoke tokens or disable the endpoint from the connector Dashboard at any time.

AIsuru Browser

Source: AIsuru gateway on the playwright-mcp engine (Microsoft, Apache-2.0) hosted next to the gateway · Type: web browsing on the user's behalf

Lets the agent open web pages, read their text, follow a link and fill in a form, staying within the list of sites the administrator declared. Each person gets their own browser window, with their own cookies and logins.

Data collected by the gateway

• The map of open windows: for each person browsing, the identifier of their session on the engine, the agent and connector it belongs to, the last address opened, and the times of opening and expiry. The person is identified by a key derived from their AIsuru identity, not by their email address in clear.
• The list of allowed domains, configured by the administrator.

Data NOT retained

• The content of visited pages, which passes through memory during execution and reaches the agent truncated to the first 24,000 characters. Cookies, logged-in sessions and data typed into forms live in the browser window on the engine, not in the gateway, and disappear when the window closes. The gateway takes no screenshots, downloads no files and runs no code inside pages.

Revocation

The close tool ends the window immediately and deletes the corresponding record. Windows left open close by themselves on expiry (ten minutes of inactivity), and removing the connector from the agent deletes the whole map. What a site wrote in its own logs stays theirs: processing by the sites visited is governed by their own privacy policies.

AIsuru Data Analysis

Source: npm package mcp-js-executor · Type: sandboxed JavaScript execution on JSON data

Lets the agent run JavaScript code in a sandbox over JSON data passed in the conversation. Useful for transformations, aggregations, calculations on data provided by the agent.

Data retained by the gateway

• None specific: code and JSON data are passed as call parameters and processed in memory.

Data NOT retained

• JavaScript code and JSON data are not persisted.

AIsuru Network

Source: internal Memori HTTP server memori.ai · Type: AIsuru platform integration

Lets the agent query the AIsuru platform itself for internal orchestration tasks (memori lookup, sessions, etc.).

Data retained by the gateway

• Backend / engine / tenant URLs configured by the MCP server administrator.
• Optional Giver credentials (tag + PIN) encrypted at-rest.

Data NOT retained

• Conversation contents with the agent are never duplicated on this MCP.

AIsuru Persistence

Source: internal Memori gateway service (MongoDB) · Type: per-agent personal database

Personal MongoDB database for each agent, used by the agent itself to store memories, states, preferences, and structured data.

Data retained by the gateway

• Arbitrary data the agent chooses to save (shape and content determined by the agent's prompt).

Data NOT retained

• No additional application-level encryption beyond the standard MongoDB Atlas Memori cluster.

Revocation

The agent admin can reset an agent's database by deleting the linked Persistence McpServer.

AIsuru Scheduler

Source: internal Memori gateway service · Type: recurring task scheduler

Lets the agent schedule recurring tasks executed autonomously in the background (e.g., "every morning at 6 send me a summary of my emails"). Tasks are user-isolated.

Data retained by the gateway

• Task definition (description, cron, prompt to send to the agent at the next execution).
• Reference to the agent and the chat user who created the schedule.
• Optional encrypted memori_password at-rest (only needed for PRIVATE memori agents).

Data NOT retained

• Execution results of the agent are not duplicated by the Scheduler.

Revocation

The user can delete scheduled tasks via the "cancel schedule" command in chat or via the gateway API.

AIsuru Skills

Source: AIsuru gateway (sandboxed execution) · Type: executable skills with code

Lets the agent run skills prepared by the organisation (instructions plus code) inside an isolated sandbox, with a workspace that persists between runs. It is the connector that stores the most data, because it keeps a record of each run.

Data collected by the gateway

• The skills: name, description, instructions, attachments and code, uploaded by whoever is entitled to manage them.
• The email addresses of the people authorised to create and edit skills (SKILL_EDITORS parameter).
• The workspace: files produced or downloaded during runs, kept for the agent until an administrator removes them.
• The run log: code executed, outcome, duration, error messages, produced output and who asked for the run. If a skill processes personal data, that data can appear in the recorded output.

Data NOT retained

• The conversation with the agent. The sandbox has no internet access other than to explicitly allowed destinations.

Revocation

From the Skills Dashboard you can delete individual skills, the workspace files and the run history. Removing the connector from the agent deletes everything.

AIsuru Document Archive

Source: AIsuru gateway (gateway database) · Type: document archive per company and per agent

Keeps the documents that people at a company upload on an agent, so that the agent can use them in every conversation with them. Each company has a separate archive on each agent, and never sees another company's documents. The company of whoever is writing comes from their login, never from what is written in the chat. If the connector provides for it, each person also has a personal archive that only they can see.

Data collected by the gateway

• The documents: the original file and the extracted text, split into passages for search, in the gateway database. The file is not sent to external services, with one exception: scanned pages, up to 20 per document, and PNG and JPG images are transcribed by the agent's model. To do so, the page image goes through AIsuru as an attachment of the reading, and the reading stays in the agent's conversation history. Beyond 20 pages, and with the parameter LETTURA_SCANSIONI = ocr, scans are read with OCR inside the gateway and never leave it.
• Each document's card: tags and description written by whoever uploads it and, if the automatic description is on (DESCRIZIONE_AI parameter), description, type and keywords written by the agent's model. To write them the gateway passes the beginning of the text to the same model that reads the documents when the agent answers in the chat.
• Who uploaded them and when: the person's verified email.
• The company registry, if the connector uses it: each person's email and the code and name of their company, written by the system of whoever integrates the agent.
• The record of operations: uploads, deletions and openings of the original, with the document name, the email and the date. It is kept for 400 days.

Data NOT retained

• The conversation with the agent. The gateway's execution log contains neither the text of the documents nor the links to open them.

Retention and deletion

A document stays until someone at the company deletes it, from the panel or by asking the agent, or until the expiry set in the connector. Deleting it removes the file and the text; the record keeps that it existed, with its name and who deleted it. Removing the connector from the agent deletes the archives of every company on that agent.

AIsuru Vibe Coder

Source: internal Memori gateway service · Type: conversational agent building (functions, views, forms, connectors)

Lets the agent's owner build and modify it by talking: create functions and contents, activate connectors, generate views and forms backed by real data, define the application's governance rules. Modification actions are restricted to the verified owner/giver.

Data retained by the gateway

• Application profile (name, purpose, data-visibility rules, brand) defined during onboarding.
• Definitions of generated views/forms and registered endpoints; endpoint credentials (tokens, username/password) are encrypted at-rest.
• Audit trail of agent modifications (who, what, when) and technical view-invocation statistics.
• Files uploaded by users through generated forms (images/documents): stored on cloud storage and referenced by unguessable public URLs.

Data NOT retained

• Conversation contents are not duplicated; application data submitted through forms lives in the agent's Persistence database (see "MCP Persistence").

Revocation

The owner can disable views and endpoints from the connector Dashboard, or remove the Vibe Coder McpServer from the agent.

Bybit Trading

Source: OFFICIAL Bybit npm package bybit-official-trading-server · Type: crypto exchange (market data, account, orders) via V5 APIs

Lets the agent query real-time market data (even without credentials) and, with configured credentials, read balances/positions and manage orders. Orders always require the user's explicit confirmation. Supports the testnet and Bybit EU accounts.

Data retained by the gateway

• Bybit API Key and API Secret configured by the administrator, encrypted at-rest.
• Configuration flags (testnet, EU host).

Data NOT retained

• Market data, balances, positions and order outcomes transit in memory during execution and are not stored by the gateway.

Revocation

Remove the McpServer from the panel, or revoke the API key from Bybit (API Management). Restricting the key to the gateway's egress IPs is recommended.

ClickUp

Source: OFFICIAL ClickUp remote service mcp.clickup.com · Type: tasks, lists, docs, chat and time tracking

Lets the agent work with the user's ClickUp workspace (tasks, lists, folders, spaces, docs, comments, chat, time tracking). Each user connects with their own account via OAuth: the agent operates with their permissions. ClickUp exposes no deletion tools.

Data retained by the gateway

• ClickUp OAuth access token (and refresh token, if issued) for each connected user, encrypted at-rest.
• OAuth client ID obtained through Dynamic Client Registration.

Data NOT retained

• The contents of tasks, docs, comments and chat are not stored by the gateway: they transit in memory during execution. Processing on ClickUp's side is governed by ClickUp's privacy policy (third party).

Revocation

From the connector Dashboard (remove the connected account), or from ClickUp → Settings → Apps by revoking the authorisation.

Cloudflare MCP Server

Source: official Cloudflare npm package @cloudflare/mcp-server-cloudflare (GitHub) · Type: Workers/KV/R2/D1 management

Lets the agent manage the user's Cloudflare resources (Workers, KV, R2, D1, Durable Objects, Queues, Workers AI, Workflows).

Data retained by the gateway

• Cloudflare Account ID.
• Cloudflare API token (encrypted at-rest).

Data NOT retained

• Memori does not retain copies of Workers code, KV/R2/D1 data, or other content managed via the connector.

Revocation

To revoke: remove the API token from the MCP server panel, or revoke the token directly from your Cloudflare account.

CRM in Cloud

Source: npm package @memori.ai/mcp-crmincloud · Type: CRM (contacts, companies, opportunities)

Lets the agent read and update data in TeamSystem's CRM in Cloud: contacts, companies, opportunities and activities.

Data collected by the gateway

• CRM API key and address (configured by the administrator, encrypted at rest).

Data NOT retained

• Contact and company data is not saved by the gateway: it passes through memory during execution. Processing on the CRM in Cloud side is governed by TeamSystem's privacy policy (third party).

Revocation

By removing the connector from the agent, or by revoking the API key in CRM in Cloud.

Excel MCP Server

Source: PyPI package excel-mcp-server (haris-musa) · Type: spreadsheet creation/editing (.xlsx)

Lets the agent create and edit Excel files (cells, formulas, charts, tables, formatting). Files are processed in the gateway workspace. Available to authenticated users only.

Data collected by the gateway

• The Excel files generated by the agent, stored temporarily in the gateway workspace and on an encrypted storage bucket (provider-side at-rest encryption), isolated per individual user (key derived from the authenticated identity, in hashed form: no personal data in paths).
• No credentials: this MCP does not connect to external accounts.
• The file download link is a URL with an unguessable random key.

Data NOT retained

• Files are automatically deleted within 24 hours (from both the workspace and the storage); after that, the download link is no longer valid.
• Any templates provided via URL are downloaded only temporarily, under the same retention rules.
• Conversation content is not duplicated beyond the files generated at the user's request.

Revocation

Files expire and are deleted automatically within 24 hours. There are no persistent credentials to revoke.

Fetch Url

Source: npm package @tokenizin/mcp-npx-fetch · Type: download content from public URLs

Lets the agent fetch content from a URL and convert it to text, HTML, JSON, or Markdown.

Data retained by the gateway

• No persistent credentials: the URL is provided by the agent at execution time.

Data NOT retained

• Downloaded content is not saved by the gateway.

Firebird MCP Server

Source: npm package mcp-firebird (PuroDelphi) · Type: SQL queries on Firebird databases (including legacy line-of-business systems)

Lets the agent query via SQL and explore the schema of a Firebird database, connecting over the native Firebird protocol (no ODBC). A read-only user on dedicated views is recommended.

Data retained by the gateway

• Firebird host, port, database, user and optional role (configured by the admin).
• Firebird password encrypted at-rest.

Data NOT retained

• Query results and table/view data are not saved by the gateway.

Revocation

The admin can remove the McpServer from the panel (clears the credentials) or revoke the Firebird user on the database.

Free Image Generator

Source: internal Memori gateway using the public service pollinations.ai · Type: image/text/audio generation

Generates images, text, or audio via the free pollinations.ai API. Supports generation, editing, and image-to-image.

Data retained by the gateway

• Optional Pollinations token (encrypted at-rest), default model, default dimensions.
• Generated images, for 30 days: they back the signed link the agent shows them with, which expires together with them.

Data NOT retained

• Prompts sent, and generated text and audio, flow through the gateway but are not stored on the Memori side.
• Handling of prompts and media is governed by pollinations.ai (third party) — see their website.

Gamma

Source: OFFICIAL Gamma remote service mcp.gamma.app · Type: generated presentations, documents and webpages

Lets the agent create presentations, documents, webpages and images in Gamma, read back the ones already in the workspace, and export them.

The connector can be used in three ways, chosen by whoever configures it: with the workspace key (everyone works in the same space), per user (each person connects their own Gamma account) or hybrid (whoever connected their own account uses it, everyone else uses the key). What the gateway retains depends on which mode is active.

Data collected by the gateway

• The workspace's Gamma API key, when used (configured by the administrator, encrypted at rest) and sent to Gamma in the X-API-KEY header.
• In per-user and hybrid mode: the OAuth access and refresh token of each person who connected their own account, encrypted at rest, together with that person's identifier in the AIsuru system, the token expiry, the granted permissions and the date of last use. Not the email address: the identifier is the one from the verified session.
• The OAuth client ID obtained through dynamic client registration, which is not a secret.

Data NOT retained

• Generated content lives in the Gamma workspace, not in the gateway: here it passes through memory during execution. The viewer analytics Gamma provides contain names and email addresses of the people who opened a document: those pass through too, without being retained. In per-user and hybrid mode the agent only sees the Gammas that person has access to, and credits are drawn from their workspace. Processing on the Gamma side is governed by Gamma's privacy policy (third party, United States).

Revocation

Each person can have their own connection removed from the connector's Dashboard, and revoke the authorisation from their own Gamma account. Removing the connector from the agent deletes every retained token. The workspace key is revoked in Gamma → Settings → API Keys.

Ghost CMS

Source: npm package @fanyangmeng/ghost-mcp · Type: Ghost blog management (posts, pages, tags, members) via Admin API

Lets the agent create, edit and publish posts and pages, manage tags, members and settings of a Ghost CMS site. The agent asks for confirmation before publishing or deleting content.

Data retained by the gateway

• Ghost site URL and Admin API Key configured by the administrator; the Admin API Key is encrypted at-rest.

Data NOT retained

• Post/page contents and member data are not stored by the gateway.

Revocation

Remove the McpServer from the panel, or delete the Custom Integration in Ghost (Settings → Integrations), which invalidates the key.

Glacier

Source: remote service getglacier.ai (HTTP MCP) · Type: AI-native project management (Kanban board + docs)

Lets the agent read and write the user's Glacier workspace — projects, cards, docs, comments and suggestions — as a collaborator. Access reflects the permissions of the connected account.

Data retained by the gateway

• Glacier API key (starts with glc_sk_), encrypted at-rest and sent to the Glacier service as a Bearer token.

Data NOT retained

• Content of projects, cards, docs and comments is not saved by the gateway: it transits in memory during tool execution. Data handling on the Glacier side is governed by getglacier.ai (third party).

Revocation

Remove the McpServer from the panel, or revoke the API key in Glacier (Account Settings → MCP API Key).

Google Workspace MCP Server

Source: PyPI package workspace-mcp (GitHub) · Type: Docs, Sheets, Slides, Drive, Calendar, Gmail

Lets the agent operate on the user's Google account on the user's own instructions: read and send email, manage Calendar events, find and read documents in Drive, create/read/modify Docs/Sheets/Slides.

Data retained by the gateway

• Google OAuth refresh token (encrypted at-rest), used to obtain short-lived access tokens (1 hour).
• Email of the connected Google account (for identity binding).

OAuth scopes requested

The consent screen presented by Google always lists the permissions this installation actually requests, which may be a subset of the following:

• openid, email, profile — identity
• drive.file — files created or explicitly opened through the app
• drive.readonly — read-only access to the user's Drive, so the agent can find and read documents the user asks about
• documents / spreadsheets / presentations — read/write Docs/Sheets/Slides
• calendar — read/write Calendar events
• gmail.readonly — read messages, so the agent can search, summarise and answer questions about the user's email
• gmail.send — send messages on the user's behalf; this permission grants no read access

The agent is never given the ability to delete email or to delete Drive files: neither gmail.modify nor the full drive scope is ever requested.

Data NOT retained

• Message bodies and attachments, Drive file contents, Doc/Sheet/Slide contents and Calendar events pass through the Service only to fulfil the request in progress, and are not stored by the gateway.
• No copy or index of the mailbox or of the Drive is ever built.

Use of Google user data with AI models

To carry out what the user asks in the conversation (for example "summarise the email that arrived today" or "what does this contract say"), the relevant content is passed to the language model behind the agent, for the sole purpose of producing the answer shown to that same user.

• Google user data is never used to develop, improve or train generalized or non-personalized AI or ML models, whether ours or third parties'.
• Google user data is never used for advertising or profiling, is never sold, and is never transferred to anyone other than the model providers strictly necessary to produce the answer requested by the user.
• No human being reads Google user data, except with the user's explicit consent (for instance on a support request the user has opened), where strictly necessary for security purposes, or where required by law.

How Google user data is secured

• Every call to Google APIs and every connection between the user, the Service and the model provider travels over TLS 1.2 or above. Google user data is never transmitted in clear text.
• The refresh token is encrypted at-rest (Lockbox, AES-256-GCM, key managed by Memori and held outside the database), and access tokens are short-lived (1 hour) and kept in memory only.
• Each user's Google credentials are bound to that user's authenticated identity: one user's token can never be used to answer another user, nor by an agent the user has not connected it to.
• Message bodies, file contents and calendar entries are not written to disk, to logs or to any index: what is logged of a tool call is the operation and its outcome, not the content.
• Access to production systems is restricted to named administrators over authenticated channels, and no member of staff can read Google user data through the Service (see above).

Retention and deletion

The refresh token is retained until the user revokes it, and is deleted immediately when the Google account is disconnected or when the connector is removed from the agent. Google content is never retained beyond the request being served. A user may request deletion of their data at any time by writing to privacy@memori.ai or to support at ccare@memori.ai; we confirm the deletion in writing.

Google API Services User Data Policy compliance

Use of data received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

Revocation

From the Service: "Disconnect Google" button on the agent page, or the "disconnect Google" command in chat. From Google: myaccount.google.com/permissions → remove authorization for "AIsuru".

HubSpot

Source: OFFICIAL HubSpot npm package @hubspot/mcp-server · Type: CRM (contacts, companies, deals, activities)

Lets the agent read and update HubSpot CRM objects, within the scopes granted to the private app.

Data collected by the gateway

• HubSpot private app token (configured by the administrator, encrypted at rest).

Data NOT retained

• CRM data is not saved by the gateway: it passes through memory during execution. What the agent can see is decided by the private app's scopes. Processing on the HubSpot side is governed by HubSpot's privacy policy (third party).

Revocation

By removing the connector from the agent, or by deleting the private app in HubSpot → Settings → Integrations.

IBM Db2 for i (AS/400)

Source: OFFICIAL IBM npm package @ibm/ibmi-mcp-server · Type: SQL queries on Db2 for i

Lets the agent explore libraries and tables of an IBM i (AS/400) system and run read-only queries on the data.

Data collected by the gateway

• Host, port, user profile and password of the IBM i system (configured by the administrator, encrypted at rest).

Data NOT retained

• Query results are not saved by the gateway. What the agent can read is decided by the user profile's rights on the IBM i system.

Revocation

By removing the connector from the agent, or by disabling the user profile on the IBM i system.

Linear

Source: OFFICIAL Linear remote service mcp.linear.app · Type: issue tracking and project management

Lets the agent read and write issues, projects, cycles and comments of the connected Linear workspace, with the permissions of the configured credential.

Data retained by the gateway

• Linear API key (or OAuth access token) configured by the administrator, encrypted at-rest and sent to Linear as a Bearer token.

Data NOT retained

• The contents of issues, projects and comments are not stored by the gateway. Processing on Linear's side is governed by Linear's privacy policy (third party).

Revocation

Remove the McpServer from the panel, or revoke the API key from Linear's settings.

Microsoft Dynamics 365

Source: npm package mcp-dataverse (codeurali) · Type: Microsoft Dynamics 365 / Dataverse via Web API v9.2

Lets the agent read and write Microsoft Dynamics 365 data (Sales, Customer Service, Field Service, Marketing, Project Operations) and any Power App built on Microsoft Dataverse. Authentication is application-level (Entra ID service principal, OAuth 2.0 Client Credentials): all users of the agent operate with the permissions of the security role assigned to the App in the Dataverse environment.

Data retained by the gateway

• The Dataverse environment URL (e.g. https://org…crm.dynamics.com).
• Microsoft Tenant ID, Client ID and Client Secret of the Azure App registered by the administrator: the Client Secret is encrypted at rest (Lockbox AES-256-GCM).

Data NOT retained

• The content of Dataverse records (accounts, contacts, leads, opportunities, cases, etc.) is not stored by the gateway: it only transits in memory during the execution of a single tool call.
• No application-level log of read/written data is kept beyond the anonymised technical trace required to operate the service.

Revocation

The agent administrator can remove the McpServer from the panel (clears the credentials). Alternatively, the Client Secret can be revoked from the Azure portal (Microsoft Entra ID → App registrations → your app → Certificates & secrets), or the Application User can be removed from the environment in the Power Platform admin center.

Microsoft Loop

Source: npm package mcp-ms-loop · Type: Microsoft Loop (.loop / .fluid) files via Microsoft Graph

Lets the agent access Microsoft Loop files via Microsoft Graph API (OAuth 2.0 Client Credentials).

Data retained by the gateway

• Microsoft Tenant ID, Client ID, Client Secret of the Azure app registered by the admin (encrypted at-rest).

Data NOT retained

• Content of Loop files is not saved by the gateway.

Microsoft SQL Server MCP

Source: npm package @donggyunryu/mcp-sql · Type: SQL queries + schema management on MS SQL Server / Azure SQL

Lets the agent run SQL queries and manage schema on a Microsoft SQL Server or Azure SQL database.

Data retained by the gateway

• Database host, name, user, password, port (configured by the admin, encrypted at-rest).

Data NOT retained

• Query results and table data are not saved by the gateway.

MongoDB MCP Server

Source: internal Memori HTTP server memori.ai · Type: MongoDB connection

Lets the agent operate on external MongoDB databases configured by the admin (read, write, aggregations).

Data retained by the gateway

• MongoDB connection string and database name (configured by the admin, encrypted at-rest).

Data NOT retained

• MongoDB documents read/written are not saved by the gateway.

Monday.com MCP Server

Source: official Monday.com npm package @mondaydotcomorg/monday-api-mcp (GitHub) · Type: Monday.com integration

Lets the agent interact with the user's Monday.com boards, items, columns, and workspaces.

Data retained by the gateway

• User's Monday.com API token (encrypted at-rest).
• Workspace name.

Data NOT retained

• Contents of Monday.com boards, items, and other resources are not saved by the gateway.

MySQL MCP Server

Source: npm package @benborla29/mcp-server-mysql · Type: SQL queries on MySQL

Lets the agent inspect schema, run SQL queries, and (optionally) perform write operations on a MySQL database.

Data retained by the gateway

• MySQL host, port, user, password, and database name (configured by the admin, encrypted at-rest).

Data NOT retained

• Query results are not saved by the gateway.

n8n Workflow Server

Source: npm package supergateway + external n8n workflow configured by the user · Type: MCP bridge to a self-hosted n8n workflow

Exposes one or more of the user's self-hosted n8n workflows to the agent as MCP tools, via a supergateway SSE bridge.

Data retained by the gateway

• n8n workflow URL.
• Optional auth token for the workflow (encrypted at-rest).

Data NOT retained

• Content transferred between agent and n8n workflow is not saved by the gateway. Data handling on the n8n side is the responsibility of the workflow admin.

OAuth / API Connector

Source: generic connector developed by Memori (internal gateway) · Type: HTTP/REST integration with configurable authentication (static token or OAuth2)

Generic connector that exposes to the agent one or more HTTP/REST endpoints configured by the administrator (e.g. corporate APIs, catalogs, LMS such as Docebo, token-protected webhooks). The gateway composes the calls, injects authentication and routes the responses.

Data retained by the gateway

• Endpoint definitions (method, URL, parameters) and the optional OpenAPI/Swagger spec URL, configured by the administrator.
• Configured authentication credentials, encrypted at-rest: static Bearer token, OAuth2 client_id/client_secret, username/password, refresh token, or PEM private key for JWT assertion (RS256). Obtained access tokens are kept in a temporary cache and refreshed automatically on expiry.

Data NOT retained

• The content of API requests and responses transits only in memory during a single tool execution and is not persisted by the gateway (very large responses are truncated).

Revocation

The administrator can remove the McpServer from the panel (clears endpoints and credentials) or revoke/rotate the credentials (client secret, token) at the API provider.

Oracle Database (ORDS)

Source: OFFICIAL MCP endpoint of Oracle REST Data Services ORDS 26.2+ · Type: SQL queries on Oracle Database

Lets the agent explore an Oracle database schema and run read queries, through the MCP endpoint exposed by ORDS.

Data collected by the gateway

• Address of the ORDS MCP endpoint and the JWT token used to authenticate (configured by the administrator, encrypted at rest).

Data NOT retained

• Query results are not saved by the gateway. What the agent can read is decided by the permissions of the user the JWT was issued for: the gateway carries the credential, it does not widen any rights.

Revocation

By removing the connector from the agent, revoking the token at the identity provider, or disabling feature.mcp on ORDS.

Outlook MCP Server

Source: npm package @mcp-z/mcp-outlook · Type: Microsoft Outlook / Exchange via Microsoft Graph

Lets the agent manage the user's Outlook mailbox: read mail, search, download attachments (with text extraction from PDF/DOCX/XLSX, and visual recognition via AIsuru Vision for images/receipts), send mail, manage categories/labels.

Data retained by the gateway

• Microsoft refresh token (encrypted at-rest), used to obtain short-lived access tokens.
• UPN of the connected Outlook mailbox (for identity binding).
• Microsoft Tenant ID and Client ID of the admin's Azure app.

OAuth scopes requested

• User.Read, openid, profile, offline_access — identity + refresh
• Mail.ReadWrite — read/modify mail and attachments
• Mail.Send — send mail
• MailboxSettings.ReadWrite — manage categories/rules
• Mail.ReadWrite.Shared / Mail.Send.Shared — ONLY when the administrator enables shared mailboxes (per-server opt-in): access to shared mailboxes the user already has an Exchange delegation for

Data NOT retained

• Mail content, attachments, and bodies are processed in memory during tool execution and are not persisted.

Revocation

From the Service: "Disconnect" button on the Outlook agent page. From Microsoft: myaccount.microsoft.com → "Apps and services" → remove the AIsuru app authorization.

PostgreSQL MCP Server

Source: npm package @henkey/postgres-mcp-server (GitHub) · Type: PostgreSQL query and schema management

Lets the agent run SELECT/INSERT/UPDATE/DELETE queries, manage schema, indexes, RLS, users, and diagnose issues on a PostgreSQL database.

Data retained by the gateway

• PostgreSQL connection string (encrypted at-rest).
• Optional SSL/TLS certificate paths if configured.

Data NOT retained

• Table data and query results are not saved by the gateway.

Salesforce MCP Server

Source: npm package @tsmztech/mcp-server-salesforce · Type: read / write on Salesforce

Lets the agent access Salesforce data and features through the Connected App (OAuth 2.0 Client Credentials, one «Run As» user for everyone) or, with per-person login, with the Salesforce account of whoever is chatting.

Data retained by the gateway

• Salesforce Connected App Client ID and Client Secret (encrypted at-rest).
• Instance URL, connection type and access mode.
• With per-person login: each person's Salesforce access token and refresh token, one per connector, encrypted at rest, together with the person's AIsuru identifier and the date of last use.

Data NOT retained

• Salesforce records read/written are not saved by the gateway.

Revocation

By removing the connector from the agent, or by revoking the app's access in Salesforce (for a single person, from Personal Settings → Connected Apps).

SAP S/4HANA · ECC (OData)

Source: OData services of the customer's SAP system · Type: business data (master data, orders, products)

Lets the agent search and read data in SAP through the OData services the system exposes: customers and suppliers (Business Partner), sales orders, product master. The preconfigured endpoints are read-only.

Data collected by the gateway

• Address of the OData services and the technical user's credentials — basic authentication password, or OAuth2 client ID and client secret (configured by the administrator, encrypted at rest).
• The list of enabled endpoints and services, with their permissions, which decides what the agent may ask for.
• With per-person login: each person's SAP access token and, if issued, refresh token, one per connector, encrypted at rest, together with the person's AIsuru identifier, the scopes granted by SAP and the date of last use.

Data NOT retained

• Data read from SAP is not saved by the gateway: it passes through memory during execution. Business Partner master data can contain personal data (names, addresses, contact details of natural persons): processing on the SAP side remains with the controller running that system.

Revocation

By removing the connector from the agent, or by disabling the technical user or the OData services in SAP. With per-person login, by disconnecting the person from the connector dashboard or revoking their access in SAP.

Power BI

Source: Microsoft Power BI REST APIs (api.powerbi.com) · Type: business analytics data (semantic models, reports)

Lets the agent list workspaces, semantic models and reports and query the models with read-only DAX queries, with the Microsoft account of whoever is chatting or with a service principal.

Data collected by the gateway

• Microsoft Entra tenant ID, client ID and client secret of the registered app (configured by the administrator, the secret encrypted at rest).
• With per-person login: each person's Microsoft access token and refresh token, one per connector, encrypted at rest, together with the person's AIsuru identifier and the date of last use.

Data NOT retained

• Query results and model structure are not saved by the gateway: they pass through memory during execution. Models can contain personal data (customers, employees): processing on the Power BI side remains with the controller running that tenant, under its agreements with Microsoft.

Revocation

By removing the connector from the agent. With per-person login, by revoking consent to the app from Microsoft Entra or from one's Microsoft account; with the service principal, by deleting the client secret or removing the app from the workspaces.

Sentry

Source: OFFICIAL Sentry npm package @sentry/mcp-server · Type: errors and application monitoring

Lets the agent consult errors, events and statistics of the organisation's Sentry projects.

Data collected by the gateway

• Sentry access token, host and organisation slug (configured by the administrator; the token is encrypted at rest).

Data NOT retained

• Error events are not saved by the gateway: they pass through memory during execution. Note: a Sentry event can contain personal data collected by the monitored application (email addresses, user identifiers, request payloads), depending on how that application is configured. Processing on the Sentry side is governed by Sentry's privacy policy (third party).

Revocation

By removing the connector from the agent, or by revoking the token in Sentry → Settings → Auth Tokens.

ServiceNow

Source: Table API of the customer's ServiceNow instance · Type: incidents, changes, CMDB, knowledge base

Lets the agent search and read incidents, change requests, configuration items, knowledge base articles and platform users. If the administrator leaves the creation endpoint enabled, the agent can open an incident after the user confirms.

Data collected by the gateway

• Instance address and integration user credentials (configured by the administrator, encrypted at rest).
• The list of enabled endpoints, which decides what the agent may ask for and whether it may write.

Data NOT retained

• Record content is not saved by the gateway: it passes through memory during execution. ServiceNow records contain personal data (who reported, who is assigned, work notes): what the agent can see is decided by the integration user's ACLs. If incident creation is enabled, the ticket data is written into the customer's instance.

Revocation

By removing the connector from the agent, or by deactivating the integration user in ServiceNow.

ServiceNow — Projects and portfolio

Source: Table API of the customer's ServiceNow instance · Type: projects, tasks, status reports, demands, portfolios, programs

Lets the agent search and read projects, tasks and milestones, status reports, demands, portfolios and programs of the Strategic Portfolio Management module. The connector is read-only: it does not create or modify records.

Data collected by the gateway

• Instance address and integration user credentials (configured by the administrator, encrypted at rest).
• The list of enabled endpoints, which decides what the agent may ask for.

Data NOT retained

• Record content is not saved by the gateway: it passes through memory during execution. Project records contain personal data (managers, task assignees, authors and comments of status reports) and may contain financial data: what the agent can see is decided by the integration user's ACLs.

Revocation

By removing the connector from the agent, or by deactivating the integration user in ServiceNow.

SharePoint MCP Server

Source: npm package developed by Memori @memori.ai/mcp-sharepoint · Type: SharePoint / Microsoft Graph

Lets the agent access SharePoint documents via Microsoft Graph API. Two modes: application (OAuth 2.0 Client Credentials: all users operate with the app's permissions) or corporate login (delegated OAuth: each user connects with their own Microsoft account and sees only what they can access).

Data retained by the gateway

• Microsoft Tenant ID, Client ID, Client Secret (or certificate with private key) of the Azure app registered by the admin (encrypted at-rest).
• Site ID, Drive ID, List ID configured.
• In corporate-login mode: Microsoft access/refresh tokens for each connected user, encrypted at-rest, with the account UPN (delegated scopes Sites.ReadWrite.All, Files.ReadWrite.All, User.Read).

Data NOT retained

• SharePoint document contents are not permanently stored by the gateway; for accessed files a temporary copy with a signed, expiring download link (24 hours) may be generated.

Revocation

Corporate-login mode: from the connector Dashboard (remove the account) or from myaccount.microsoft.com. Application mode: revoke the Client Secret/certificate on Azure.

Snowflake

Source: Snowflake-managed MCP server, inside the customer's account (documentation) · Type: business data (Cortex Agent, Cortex Analyst, Cortex Search, SQL)

Lets the agent use the tools the customer defined in their own Snowflake MCP server — questions to a Cortex Agent, searches, SQL queries, functions and procedures — with the privileges of the role bound to the configured token.

Data retained by the gateway

• Snowflake Programmatic Access Token configured by the administrator, encrypted at-rest and sent to Snowflake as a Bearer token.
• Account identifier, database, schema and MCP server name, which make up the server address.
• In the execution log (30 days): the tool called, the argument names, the outcome with any Snowflake error message, and the size of the response.

Data NOT retained

• Questions, SQL queries and the data returned by Snowflake: they transit in memory during tool execution and are not stored by the gateway, not even in the execution log. The data stays in the customer's Snowflake account, and its processing is governed by the customer's agreement with Snowflake (third party).

Revocation

Remove the McpServer from the panel, or revoke the token in Snowflake (ALTER USER ... REMOVE PAT ..., or from Snowsight → Users & roles → Programmatic access tokens).

Matomo

Source: MCP server of the official MCP Server plugin, in the customer's Matomo (Cloud or on-premise) (documentation) · Type: aggregated website statistics (visits, pages, referrers, goals)

Lets the agent read the Matomo reports of the sites the Matomo user bound to the configured token can access.

Data retained by the gateway

• Matomo token_auth configured by the administrator, encrypted at-rest and sent to Matomo as a Bearer token.
• Address of the Matomo installation, from which the MCP endpoint is built.
• In the execution log (30 days): the tool called, the argument names, the outcome with any Matomo error message, and the size of the response.

Data NOT retained

• The reports and statistics returned by Matomo: they transit in memory during tool execution and are not stored by the gateway, not even in the execution log. The data stays in the customer's Matomo, and its processing is governed by the customer's configuration and agreement with Matomo (third party).

Revocation

Remove the McpServer from the panel, or delete the token in Matomo (Personal settings → Security → Auth tokens) or turn the MCP server off (Administration → System → General settings → MCP Server).

Solid Pod

Source: npm package @memori.ai/mcp-solid-pod · Type: personal data store (Solid)

Lets the agent read and, when enabled, write documents inside a Solid Pod: a personal data store that stays under the control of whoever owns it.

Data collected by the gateway

• Pod address, WebID, OIDC issuer, client ID and client secret (configured by the administrator, encrypted at rest).
• Whether the connector is read-only.

Data NOT retained

• The content of the Pod's documents is not saved by the gateway: it passes through memory during execution. The Pod stays in its owner's domain, which is the whole point of the Solid model.

Revocation

By removing the connector from the agent, or by revoking the client's authorisation at the Pod provider.

Stripe — pagamenti e abbonamenti

Source: AIsuru gateway on Stripe's API stripe.com · Type: payments, subscriptions and access entitlements

Lets an agent charge for access to its own features: it opens Stripe's hosted checkout, receives the signed payment confirmation and keeps track of who is entitled to use the agent. It is the second connector that retains data, and that data is personal.

Data collected by the gateway

• Stripe secret key and webhook secret (configured by the administrator, encrypted at rest).
• Each end user's access entitlement: email address and AIsuru user identifier, subscription status, plan, end of the paid period, number of free actions used, and the Stripe identifiers of the customer, the subscription and the last checkout.
• The log of events received from Stripe (type, outcome, time), so the same payment is not processed twice.

Data NOT retained

• No payment data. Card number, expiry and security code never pass through the gateway: payment happens on Stripe-hosted pages. Processing on the Stripe side is governed by Stripe's privacy policy (third party).

Revocation

The agent's owner can revoke an entitlement from the connector's Dashboard; removing the connector deletes the entitlements and the event history. Deletion of data held by Stripe must be requested from Stripe.

Wikipedia

Source: npm package @shelm/wikipedia-mcp-server · Type: search / read on Wikipedia (public)

Lets the agent query Wikipedia and use its public content.

Data retained by the gateway

• No credentials (Wikipedia is public and requires no authentication).

Data NOT retained

• Queries and articles read are not saved by the gateway.

Word MCP Server

Source: PyPI package docx-mcp-server (SecurityRonin) · Type: document creation/editing (.docx)

Lets the agent create and edit Word documents (paragraphs, tables, headers, tracked changes, comments), optionally starting from a company template. Files are processed in the gateway workspace. Available to authenticated users only.

Data collected by the gateway

• The Word documents generated by the agent, stored temporarily in the gateway workspace and on an encrypted storage bucket (provider-side at-rest encryption), isolated per individual user (key derived from the authenticated identity, in hashed form).
• No credentials: this MCP does not connect to external accounts.
• The file download link is a URL with an unguessable random key.

Data NOT retained

• Files are automatically deleted within 24 hours (workspace and storage); after that, the download link is no longer valid.
• Any templates (.docx/.dotx) provided via URL are downloaded only temporarily, under the same retention rules.
• Conversation content is not duplicated beyond the files generated at the user's request.

Revocation

Files expire and are deleted automatically within 24 hours. There are no persistent credentials to revoke.

Work IQ (Microsoft 365)

Source: OFFICIAL Microsoft remote service workiq.svc.cloud.microsoft · Type: Microsoft 365 work intelligence (mail, calendar, Teams, documents, Copilot)

Lets the agent access the user's Microsoft 365 work data (email, events, Teams chats and channels, documents, people) and query Microsoft 365 Copilot in natural language. Each user connects with their own corporate account: the agent sees only what the user can see. Requires a Microsoft 365 Copilot license and tenant admin consent.

Data retained by the gateway

• Microsoft access and refresh tokens for each connected user, encrypted at-rest; account UPN for identity binding.
• Microsoft Tenant ID, Client ID (and optional Client Secret, encrypted) of the admin's App Registration.

OAuth scopes requested

• openid, profile, offline_access — identity + refresh
• WorkIQAgent.Ask (api://workiq.svc.cloud.microsoft) — querying Work IQ on the user's behalf (admin consent required)

Data NOT retained

• The contents of email, events, chats, documents and Copilot answers transit in memory during tool execution and are not persisted by the gateway.

Revocation

From the connector Dashboard (remove the connected account), or from Microsoft: myaccount.microsoft.com → "Apps and services". The administrator can revoke admin consent from the App Registration in Entra ID.

Workday

Source: REST APIs of the customer's Workday tenant · Type: HR data (workers, organisations)

Lets the agent search workers, read a person's profile, walk the management chain and list organisations. The preconfigured endpoints are read-only.

Data collected by the gateway

• Tenant API address, token endpoint, client ID and client secret of the Registered API Client (configured by the administrator, encrypted at rest).
• The list of enabled endpoints, which decides what the agent may ask for.

Data NOT retained

• Personnel data is not saved by the gateway: it passes through memory during execution. This is personal data about employees and contingent workers, and what the agent can see is decided by the security domains granted to the integration user in Workday: that is the lever to use, not the agent's instructions. Processing on the Workday side remains with the controller running that tenant.

Revocation

By removing the connector from the agent, or by deactivating the Registered API Client or the security domains in Workday.

Zapier MCP Server

Source: npm package supergateway + Zapier webhook configured by the user · Type: MCP bridge to Zapier automations

Exposes the user's Zapier automations to the agent as MCP tools, via a supergateway SSE bridge.

Data retained by the gateway

• Zapier webhook URL.
• Optional auth token for the webhook (encrypted at-rest).

Data NOT retained

• Content transferred between agent and Zapier is not saved by the gateway. Data handling on the Zapier side is the responsibility of the user who set up the webhook.