Connect your AI coding assistant to Reeve

The quickest way#

Open Reeve's page (http://reeve.localhost:18383/) and its Connect your AI to Reeve card. It has the setup for each of the common assistants and editors, already filled in with this PC's own paths, each with Copy. Paste it where the card says, and restart the assistant.

The card folds to Your AI is connected to Reeve once an assistant has connected, naming the assistants and when one last used Reeve. Open it again to connect another.

What the setup is#

Reeve's MCP server runs over stdio. Your assistant starts it with Node and Reeve's command line:

C:\Users\you\.manor\node\node.exe C:\Users\you\.reeve\app\src\cli.ts mcp

The node is the one Castellan installed. These paths stay the same when Reeve updates, so the setup never needs changing.

Assistants that keep their MCP servers in a JSON file under mcpServers take this block:

{
  "mcpServers": {
    "reeve": {
      "command": "C:\\Users\\you\\.manor\\node\\node.exe",
      "args": ["C:\\Users\\you\\.reeve\\app\\src\\cli.ts", "mcp"]
    }
  }
}

VS Code (.vscode\mcp.json in a workspace, or MCP: Open User Configuration) takes its own shape:

{
  "servers": {
    "reeve": {
      "type": "stdio",
      "command": "C:\\Users\\you\\.manor\\node\\node.exe",
      "args": ["C:\\Users\\you\\.reeve\\app\\src\\cli.ts", "mcp"]
    }
  }
}

Assistants configured in TOML:

[mcp_servers.reeve]
command = 'C:\Users\you\.manor\node\node.exe'
args = ['C:\Users\you\.reeve\app\src\cli.ts', 'mcp']
tool_timeout_sec = 600

An assistant with its own mcp add command takes the same command line after it, at user scope, so every project has Reeve.

Tell your assistant to use it#

On connecting, Reeve tells the assistant what each tool is for, and that its answers are leads to check. Assistants still forget it, and read every file themselves. Put this line in the assistant's own instructions file (AGENTS.md, or whatever your assistant reads at the start of a session):

Reeve is the local AI layer on this PC (MCP server "reeve"). Before reading large files, logs or a library's docs yourself, ask Reeve: search to find code, triage for a long build or test log, summarize for a big file, docs for a dependency's pinned docs. At the start of a session, call Reeve's digest tool and mention any alert for this repository. Treat its answers as leads to check, not facts.

The card on Reeve's page has it too, with Copy.

Reminders

For the command-line assistant whose hooks Reeve supports, the card offers Reeve's reminders. They remind the assistant of Reeve's tools at the moment it would use them:

• when a session starts: what each tool is for;
• after 6 searches of its own since it last used Reeve: Reeve's search;
• reading a whole file of 60 KB or more: summarize; a log: triage;
• 20 KB or more of command output: redirect it to a file, and triage it next time.

Each reminder comes at most once every 20 minutes, and only while Reeve's server is up. They never stop a tool. Turning them on also allows Reeve's read-only tools by name, so the assistant isn't kept waiting for a yes before each one; a tool that changes something would still be asked. Run once:

node %USERPROFILE%\.reeve\app\src\cli.ts hooks on

hooks off takes them out, and hooks says where they stand.

Check that it's used#

• The card on Reeve's page says which assistants have connected, and when one last used Reeve.
• Kept off the cloud grows as your assistant uses Reeve: see Kept off the cloud.
• The ai-use job looks, every hour, at the AI sessions that worked in a repository over the last day. When under 65% of them used Reeve, it raises an alert and says why: setup (most never had Reeve connected), the assistant (Reeve was there, and it forgot him), or a gap in Reeve (it ignores the reminders, or his tools fail). For a gap, ai-use --report writes an email to us, so Reeve can be improved.

Removing it#

Uninstalling Reeve, or letting it go in Castellan, takes its reminders out. The MCP entry in your assistant's settings is yours: remove the reeve entry there.