Wire any agent to a live MCP.

Register flmnt as an HTTP MCP server with Claude Code’s own “claude mcp add” command, or by hand in .mcp.json. The flmnt CLI is a separate tool that automates that and installs the full automation kit — a lifecycle hook map, 13 /flmnt-* slash commands, and tool permissions — so reading and recording memory happen on their own.

claude mcp add --transport http --scope project flmnt "http://localhost:8000/mcp"

Every connected client sees the same toolset — identical whether you came in through Claude Code, a connector, or a JSON config. Your agent reads recent context, retrieves causally across the graph, and writes back decisions, keyframes, and mistakes. The full surface:

This is the kit flmnt setup installs for you — the same way we run flmnt in our own repo. A project prompt so every session knows what to record, a full lifecycle hook map so reading and recording memory happen on their own, and a set of /flmnt-* slash commands for the on-demand writes. Here is what it wires, so you can see (and trim) it.

# flmnt memory

This repo records to the **your** workspace via MCP.

- Record a **decision** whenever you commit to an approach (and why).
- Drop a **keyframe** at the end of any working session — current state, open threads.
- Capture a **mistake** the moment a run regresses, with the cause.

Read recent keyframes before starting work. Memory persists across sessions.

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          { "type": "command", "command": "flmnt gate" }
        ]
      }
    ]
  }
}

{
  "hooks": {
    "SessionStart": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "flmnt brief" }] }
    ],
    "UserPromptSubmit": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "flmnt gate" }] }
    ],
    "Stop": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "flmnt derive --hook" }] }
    ]
  }
}