Jitsu MCP Server認証済み注目

@Jitsu

リポジトリホームページドキュメント

基本情報

トランスポート

Streamable HTTP

認証方式

OAuth

カテゴリ

データと分析

作者

Jitsu

投稿者

Soham Mehta

Jitsu MCP Server について

Jitsu is an open-source Customer Data Platform — an MIT-licensed alternative to Segment. This MCP server gives your AI assistant direct control over a Jitsu workspace, so you can run and debug data pipelines in conversation instead of clicking through the console. 25 tools cover the full workflow: • Configuration — inspect and edit destinations, streams, services, and the connections between them, including reading the JSON schema for any resource type before creating or updating it. • Syncs — trigger a sync, cancel a running one, list recent tasks, and pull task logs newest-first when something fails. • Sources — fetch a connector spec, discover available streams, verify credentials, and test a connection before wiring it into production. • Live events — query recent event-log records across incoming streams, function runs, bulker writes, and the dead-letter queue, filtered by source, time range, or search term. • Functions — execute a Jitsu function or profile builder against real data without persisting anything. • Statistics — event throughput, sync history, and profile builder metrics. Every tool carries read-only or destructive annotations, so reads run without interrupting you and anything that modifies or deletes configuration asks first. Authentication is OAuth 2.1 with dynamic client registration — there is no API key to create or paste. You approve access once and the client connects to the workspaces your Jitsu account already has. Works with Jitsu Cloud or a self-hosted instance.

#cdp #analytics#data-pipeline#etl#event-streaming#data-warehouse #clickhouse#segment-alternative#open-source

接続情報

https://use.jitsu.com/mcp

セットアップ

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

Jitsu MCP Server のツール

List the Jitsu workspaces the authenticated user can access (paginated, max 100 per page). Returns { workspaces, total, limit, offset, hasMore }; page with `offset` while `hasMore` is true. Use a workspace `id` as the `workspaceId` argument to other tools.

List configuration resources of a given type in a workspace. type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}. Use "connection" for connections (links between streams/services and destinations).

Get a single configuration resource by id. type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}.

JSON Schema describing the payload for creating/updating a resource. type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}. Use `subtype` to narrow: a destination/service kind (e.g. "postgres"), or for "connection" the connection kind ("sync", or a destination type for push connections).

Create a configuration resource. type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}. `data` must match get_resource_schema for the type. For "connection", `data` is the connection body: { fromId, toId, type: "push"|"sync", data }.

Update a configuration resource by id (merged into the existing object). type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}. For "connection", `data` is the connection body and the link is upserted by id.

Delete a configuration resource by id (soft delete). type ∈ {destination, stream, function, service, custom-image, domain, notification, connection}.

List the sources (actors) you can pass as `source` to query_events, for a workspace. Optionally narrow by `type`: "incoming" → streams; "function"/"bulker_batch"/"bulker_stream" → connections, destinations, profile builders; "dead-letter" adds the "all" sentinel.

Read recent event-log records for a workspace. type ∈ {incoming, function, bulker_batch, bulker_stream, dead-letter}. `source` filters to one actor (a stream id for "incoming"; a connection id for "function"/"bulker_*"; any actor or "all" for "dead-letter"). Omit `source` (or pass "all") to query across ALL the workspace's sources. Use list_event_sources to discover ids. Results are newest-first, capped by `limit`.

Schedule a sync (a connection of type "sync") to run immediately. Returns the new `taskId`; poll list_sync_tasks with it for status and get_sync_logs for output. If a task is already running the call fails and returns `runningTask` — pass ignoreRunning=true to cancel it and start anyway. fullSync=true drops the saved cursor and re-syncs from scratch.

Cancel a running sync task. Cancellation is asynchronous — poll list_sync_tasks until the task status becomes CANCELLED.

Up to 50 most recent sync tasks in a workspace, newest first. Filter by `syncId`, a single `taskId` (returns `task`), `status` (RUNNING, SUCCESS, PARTIAL, FAILED, CANCELLED, SKIPPED), and a `from`/`to` time window.

Log records of a sync task, newest first (default 200 records, max 5000).

JSON schema of a connector package's credentials (the `credentials` field of a service). ASYNC: the first call starts a fetch on the sync controller and returns `{ ok: false, pending: true }` — call again (every few seconds) until `ok: true` and `specs` is set, or `error` is returned. force=true bypasses the cache.

Ask the connector of an existing service to list its available streams (the catalog used to configure a sync). ASYNC: returns `{ ok: false, pending: true }` while running — call again (every few seconds) until `ok: true` and `catalog` is set. refresh=true re-fetches an already-cached catalog; force=true ignores the cache entirely.

Start an async credentials check for an existing service (source) using its stored config. Returns `{ pending: true, storageKey }` — poll get_source_check_result with the storageKey until it resolves. For destinations use test_connection instead (synchronous).

Poll the result of check_source_credentials: `{ ok: true }` on success, `{ ok: false, error }` on failure, `{ ok: false, pending: true }` while still running.

The saved incremental cursor of a sync, keyed by stream (empty object if none saved).

Delete the saved cursor of a sync so the next run re-syncs from scratch. Pass `stream` to reset a single stream, omit it to reset all. Fails while the sync is running. Returns the remaining state.

Synchronously test destination credentials via bulker. Pass `id` to test an existing destination as stored, or `config` (see get_resource_schema) to test an unsaved config — masked secrets in `config` are resolved from the object referenced by `config.id`. For services (sources) use check_source_credentials instead.

Run a transformation function against a sample event on the workspace's functions server and return the result, logs, and store mutations. `code` omitted → the stored function's draft/code is used. Nothing is persisted; events are not sent anywhere.

Run a profile-builder function against sample events and return the computed profile, logs, and store mutations. `code`/`settings` omitted → loaded from the stored profile builder. Nothing is persisted.

Event counts per period, connection, stream, destination, and status (success/dropped/error) for a workspace. Defaults to the last month with day granularity.

Number of distinct source→destination pairs with at least one successful sync task in the period (defaults to the last month).

Build progress of a profile-builder version: status (ready/building/unknown), processed/total users, and speed. `version` omitted → the current version.

概要

What is Jitsu?

Jitsu is an open-source Customer Data Platform. It collects event data from your websites, apps, and servers and delivers it to your data warehouse and to whatever other tools you use. It covers the same ground as Segment, but it's MIT-licensed and self-hostable — the whole pipeline can run inside your own infrastructure, or on Jitsu Cloud.

The MCP server ships inside the Jitsu console, so it works with either.

What you can do with it

Connect it and your assistant can run and debug pipelines in conversation instead of clicking through the console:

Debug a failing sync — list recent sync tasks, pull the logs newest-first, inspect the connection, re-run it.
Query live events — read the event log across incoming streams, function runs, bulker writes, and the dead-letter queue, filtered by source, time range, or search term.
Edit configuration — read the JSON schema for a resource type, then create or update destinations, streams, services, and connections.
Test before shipping — verify source credentials, discover available streams, and run a Jitsu function or profile builder against real data without persisting anything.

Tools

25 tools, grouped by what they touch:

Area	Tools
Configuration	`list_workspaces`, `list_resources`, `get_resource`, `get_resource_schema`, `create_resource`, `update_resource`, `delete_resource`
Syncs	`run_sync`, `cancel_sync`, `list_sync_tasks`, `get_sync_logs`, `get_sync_state`, `reset_sync_state`
Sources & connectors	`get_connector_spec`, `discover_streams`, `check_source_credentials`, `get_source_check_result`, `test_connection`
Events & functions	`list_event_sources`, `query_events`, `run_function`, `run_profile_builder`
Statistics	`get_event_stat`, `get_sync_stat`, `get_profile_builder_stats`

Every tool declares a title and the applicable readOnlyHint / destructiveHint. Thirteen are read-only, so your client can run them without a confirmation prompt; anything that modifies or deletes configuration is marked destructive and asks first.

Connecting

Endpoint: https://use.jitsu.com/mcp
Transport: Streamable HTTP
Auth: OAuth 2.1 with dynamic client registration — no API key to create or paste

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

On first connect you are sent through an OAuth approval. After that the server sees the workspaces your Jitsu account already has. Self-hosting Jitsu? The MCP server is part of the console, so your own deployment exposes it too.

Destinations Jitsu delivers to

ClickHouse, BigQuery, Snowflake, Redshift, Postgres, MySQL, S3, and GCS for warehouses and storage, plus SaaS destinations including Amplitude, Braze, HubSpot, Intercom, Mixpanel, MongoDB, PostHog, Salesforce, Segment, SendGrid, GA4, Google Ads, Google Tag Manager, Hotjar, and LogRocket.

よくある質問

Jitsu のリモート MCP サーバーとは何ですか?

Jitsu のリモート MCP サーバーは、https://use.jitsu.com/mcp にホストされた Model Context Protocol エンドポイントです。AI アシスタントはローカルに何もインストール・実行することなく接続できます。

Jitsu の MCP サーバーにはどう接続しますか?

エンドポイント https://use.jitsu.com/mcp を Claude Code、Cursor、VS Code など任意の MCP 対応クライアントに追加してください。このページのセットアップ手順で各クライアントを一度に設定できます。

Jitsu の MCP サーバーは認証が必要ですか?

はい。Jitsu は OAuth を使用します。初回接続時に MCP クライアントがブラウザを開いてサインイン・認可を行い、以降はその認証情報を再利用します。

Jitsu の MCP サーバーはどのトランスポートを使用しますか?

Jitsu は Streamable HTTP エンドポイントを公開しており、これはリモート MCP サーバーで広く使われ、主要な MCP クライアントがサポートするトランスポートです。

他のリモート MCP サーバー

AdWhispr - Research, Clone, & Launch Ads from Claude/ChatGPT

Remote MCP server that turns Claude, ChatGPT, or Cursor into a full ad workflow: research any brand's live Facebook and TikTok ads, clone the proven winners for your own brand, and launch real campaigns on Google, TikTok

Adobe for creativity

Ideate, create, and deliver with Adobe pro tools

AQL PropertyCheck — Physical check of one exact address

Check one exact Gold Coast or Brisbane City property address. Free address/coverage check and short summary; paid structured Agent JSON adds available building, street and surrounding context, with Brisbane routed journe

Adobe Journey Optimizer

Understand and troubleshoot your Journeys and Campaigns

2pr.io

Write, schedule and publish LinkedIn posts in your own voice, reply to comments and read your analytics, all from your AI assistant. Remote MCP server with OAuth sign-in.