BluetDocs

Connect Bluet to Claude

Bluet keeps what your product is for, what was decided and the rules your team wrote for agents, and gives it to the AI agents working on the product: briefed before a task, checked before a plan becomes code, debriefed after. Every answer cites the record it came from.

Connect

Claude (claude.ai, Claude Desktop): Settings → Connectors → Add custom connector, URL https://bluet.dev/mcp. You sign in to Bluet and approve access (OAuth); no token to copy.

Claude Code: the plugin adds the server, a session-start brief, a check before every commit (a blocking conflict stops the commit once, so Claude asks you), and a debrief reminder:

claude plugin marketplace add bluet-dev/bluet-plugin
claude plugin install bluet@bluet

Any MCP client, no account: npx -y bluet-mcp runs the same server locally over stdio, with memory kept in ~/.bluet. Start a product from a repository with npx bluet-mcp init.

The loop

Preflight before a task: the rules and decisions that bear on it. Check a plan or diff: conflicts with what was decided, marked blocking or warning. Debrief after work that settled something: it is proposed to the owner, never written directly. Bluet never writes a product's measure or a bet's falsifier; those are the owner's.

Example prompts

Tools

15 tools only read. 15 write to Bluet's own record (never to your repositories or other tools); those marked replaces overwrite a record in place, and Claude asks before each call.

Read-only

Write

Data and support

What Bluet stores and sends is on the privacy page. Help and bug reports: github.com/bluet-dev/bluet-plugin/issues, or from a terminal: npx bluet-mcp feedback "…".