# Jitsu — MCP server card for AI agents

Manage Jitsu data pipelines: destinations, streams, connections, functions, live events.

Registry name: `com.jitsu/mcp` · category: Data and analytics · version 1.0.0 · updated 2026-07-31

## Connect
- Remote (Claude Code): `claude mcp add --transport http jitsu https://use.jitsu.com/mcp`

## Facts
- Repository: https://github.com/jitsucom/jitsu (5,135 stars)
- Website: https://jitsu.com/docs/mcp

Before you connect a server: read its repository, and give it only the permissions and keys it needs.

## Links
- AgentGid page: https://agentgid.com/mcp/servers/com-jitsu-mcp/


## Jitsu: set up and fix your event data pipeline by chat

Jitsu collects events from your site or app, such as page views and sign-ups, and delivers them to your database or other tools. This server lets an AI assistant run that pipeline for you. It can add a destination, connect a source, watch events arrive live and fix a broken transform, without clicking through the Jitsu screens.

### Where to use it
- Send your events to a new database: Ask the assistant to add your Postgres or ClickHouse database as a destination and connect your site's events to it. It checks the login details first and then confirms that a test event landed.
- Fix a transform that throws errors: When Live Events shows a function failing, the assistant reads its code, tries a fix on a sample event and only then saves it.
- Re-run a data import from scratch: When a connector import from another service went wrong, the assistant can look at its logs, clear the saved position and start a full re-import.
- Check how many events arrive: Ask how many events each source sent this week and which destinations failed. The assistant reads the counts without changing anything.
- See what an assistant changed: Owners can read the change log filtered to changes made through this server, to review what an assistant did before.
- Pipeline changes from your build server: Keep your Jitsu setup in step with your releases by letting a scheduled job's agent apply the same configuration each time, with a personal key. (idea, not from the official pages)

### How to start
- Add the server with one command; nothing is installed.

```
claude mcp add --transport http jitsu https://use.jitsu.com/mcp
```

- On the first Jitsu tool call a browser tab opens; approve the connection once.
- Ask the assistant to list your workspaces, then work with destinations, streams and events in one of them.

### License
The Jitsu code, this server included, is free under the MIT license and can be run on your own servers. Jitsu Cloud has a free plan with 200,000 delivered events a month and paid plans above that.

### Good to know
- The assistant acts as you. A personal key carries your whole workspace access, so it can do anything you can.
- Three tools change or delete things for good. These are update, delete and the reset of an import's saved position.
- Some tools reach outside Jitsu. Credential checks contact your real data source, and test runs execute your own function code.
- Every change made this way is logged with its origin, so you can tell an assistant's changes from a person's.
- Browser sign-in does not work in scheduled jobs or build servers; those need a personal key instead.

Checked 2026-11-10 against: https://registry.modelcontextprotocol.io/v0.1/servers?search=com.jitsu/mcp&version=latest, https://github.com/jitsucom/jitsu/blob/e74ffe2bfcabca86d07aaa8384febe6b225898bc/README.md, https://github.com/jitsucom/jitsu/blob/e74ffe2bfcabca86d07aaa8384febe6b225898bc/LICENSE, https://github.com/jitsucom/jitsu/blob/e74ffe2bfcabca86d07aaa8384febe6b225898bc/webapps/console/lib/server/mcp-server/tools.ts, https://github.com/jitsucom/jitsu/blob/e74ffe2bfcabca86d07aaa8384febe6b225898bc/webapps/console/lib/server/mcp-server/index.ts, https://github.com/jitsucom/jitsu/blob/e74ffe2bfcabca86d07aaa8384febe6b225898bc/webapps/console/lib/server/mcp-server/auth.ts, https://jitsu.com/docs/mcp, https://jitsu.com/pricing

### Mistakes to avoid
- Don't: Guess the destination payload and call create_resource directly → Do: Call get_resource_schema for the type (and subtype, e.g. postgres) first, then create_resource with matching data (data must match the schema for the type.)
- Don't: Treat the first discover_streams or check_source_credentials answer as the result → Do: Poll (get_source_check_result for credentials) until the result is ready (Operations that talk to the real source are asynchronous.)
- Don't: Fix a function by calling update_resource straight away → Do: Try the fix with run_function on a sample event first, then save with update_resource (update_resource overwrites the live function; run_function persists nothing.)
- Don't: Commit the API key in .cursor/mcp.json or paste it in chat → Do: Read it from an environment variable (${JITSU_API_KEY}) in the headers block (The key carries your whole workspace access and does not have to expire.)
- Don't: Expect OAuth sign-in to work in CI → Do: Use a personal API key in an Authorization Bearer header for headless runs (The browser flow cannot run there.)

### Questions
- Do I need an API key? Not in Claude Code, Claude Desktop, Cursor or VS Code; they sign in through the browser. Only CI and other headless setups use a personal key.
- Does it work with a self-hosted Jitsu? Yes. Use your own console host; the server is at /mcp on it.
- How do I disconnect an assistant? Open the API Keys section of your account and revoke its token; it is cut off at once.
- Can the assistant test a function without breaking live data? Yes. run_function runs the code on a sample event and saves nothing.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
