Kaiten MCP with documents
kaiten-mcp
A 66-tool Kaiten MCP server, the only one in this batch supporting Kaiten documents and their inline comment threads
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Can delete cards, documents and attachments
- Reads the company audit log and data for all workspace users
Install
Manual install
git clone https://github.com/postusername/kaiten-mcp.git && cd kaiten-mcp && npm installNo ready npm package.
This is third-party code. Review the repository files before installing.
What it does
The server covers spaces, boards, columns and lanes with full CRUD, cards with filters, members and tags, comments, checklists, time logs, blockers and child cards, users, sprints, custom properties and the company audit log. A separate group of tools works with Kaiten documents: create, read, update and delete a document, upload an image and get a link for embedding, document groups, and, unlike other Kaiten servers here, inline document discussions, comment threads anchored to a specific text fragment, with creating, replying to and resolving a thread. It runs over stdio from source or as an HTTP/SSE server in Docker.
Who it is for. For teams that use Kaiten not just for boards but also for documents with fragment-level discussions.
Good fit when
- You need to read and edit Kaiten documents, not just cards
- You need discussions anchored to a specific text fragment in a document
- You need the company audit log via the list_audit_logs tool
Not a fit when
- You want a ready npm package: here you clone the repository and install dependencies manually
- You do not need documents, only boards: a simpler server will do
Example request
Open the spec document in Kaiten and start a discussion on the paragraph about environment variablesLimitations
No npm package, installed by cloning and npm install. Kaiten's rate limit is 5 requests per second, bulk operations need pauses. The inline-comment tool fails on a document that has never had a comment added through the Kaiten UI: Kaiten's API rejects or silently strips such additions, and the server detects this and suggests adding one comment manually first.
How to disable. Remove the kaiten entry from Claude Code's user or project config with claude mcp remove kaiten, or stop the Docker container.
MCP
- Transport
- stdio, sse
- Authentication
- API key
| Environment variables | |
|---|---|
| KAITEN_TOKEN required, secret | API token from Profile → Settings → API Keys |
| KAITEN_DOMAIN required | Subdomain only, without .kaiten.ru |
| MCP_HTTP_PORT | If set, the server starts as HTTP/SSE instead of stdio |
Security check
- Can delete cards, documents and attachments
- Reads the company audit log and data for all workspace users
README in short
The README lists 66 tools by section, including a separate group for documents and inline discussions, gives installation and setup instructions for Claude Code as a standalone project or via manual registration, describes the Docker HTTP/SSE mode, a card priority value table, and a notes section on practical Kaiten API quirks found by the author, including document content duplication when the UI and API edit concurrently.
FAQ
Does it support Kaiten documents?
Yes, it is the only server in this batch with full document CRUD and inline discussions on text fragments.
Why can't I add a discussion to a new document?
Kaiten's API rejects annotation marks on a document that never had a manual comment; add one comment in the Kaiten UI first.
Related
A self-hosted knowledge base with block-level references and a built-in MCP server for connecting AI agents to your notes
A CLI for every Google Workspace API with JSON output and agent skills: Drive, Gmail, Calendar, Sheets and more
Local search over Markdown notes, docs and meeting transcripts: keywords, semantic search and reranking, with an MCP server
A task manager for AI-driven development: breaks a PRD into dependent tasks and guides the agent through them via MCP or CLI