Skip to content
MCP server

CHAP Coordinator

Record human approvals, edits and overrides of AI agent work in an auditable workspace

Runs locally
Install
Platforms
CLI, self-hosted, API
Runs
Locally
Version
0.3.5
Listing updated
4 days ago

Similar MCP servers

About CHAP Coordinator

CHAP Coordinator is an MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.

CHAP is a protocol for logging human decisions over agent output, and this package exposes its Coordinator as an MCP server so an agent can create tasks, ask questions and read the audit trail. The default stdio server makes every call as one participant, agent:assistant, and refuses to act as a human or as any other participant, so approvals stay with a person in a separate desk.

The default agent tool set covers creating, updating and completing tasks, asking whisper questions, reading workspace descriptions and the audit log, and verifying the hash chain. Decision tools exist for work the agent is asked to review, but the coordinator refuses them on the agent's own work and records the attempt. Tools that choose reviewers, escalate or manage workspaces are left out. Setting review_required on a task means the agent cannot complete it alone.

Without a URL or database path it keeps everything in memory. It can forward calls to a running CHAP server, or keep workspaces in a SQLite file. Node 22 or newer is needed.

Install

Run in a terminal

claude mcp add chap -- npx -y @brightbeamai/chap-coordinator-mcp

Add to claude_desktop_config.json

{
  "mcpServers": {
    "chap": {
      "command": "npx",
      "args": [
        "-y",
        "@brightbeamai/chap-coordinator-mcp"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "chap": {
      "command": "npx",
      "args": [
        "-y",
        "@brightbeamai/chap-coordinator-mcp"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "chap": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@brightbeamai/chap-coordinator-mcp"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "chap": {
      "command": "npx",
      "args": [
        "-y",
        "@brightbeamai/chap-coordinator-mcp"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "chap": {
      "command": "npx",
      "args": [
        "-y",
        "@brightbeamai/chap-coordinator-mcp"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • CHAP_PARTICIPANTThe participant every call is made as. Defaults to agent:assistant, the agent a create-chap-app mcp-gate project joins. A call naming anyone else is refused, and so is a human: URI here, since a person decides in a desk.
  • CHAP_DB_PATHPath to a SQLite file; a leading ~ is the home directory. Without it, and without --url, the coordinator runs in memory and workspaces are lost when the client restarts.
  • CHAP_PROFILESComma-separated CHAP profiles enabled on new workspaces. Defaults to Core plus review, whisper, deliberation, handoff, control, routing, modes and audit-scitt.

Tools this server provides (11)

Taken from the project documentation; the running server may differ.

chap.task.create
Create a task.
chap.task.update
Update a task.
chap.task.complete
Complete a task.
chap.whisper.ask
Ask a whisper question.
chap.decide.approve
Approve work the agent was asked to review.
chap.decide.reject
Reject work under review.
chap.decide.override
Override a decision.
chap.review.acknowledge
Acknowledge a review.
chap.workspace.describe
Describe the workspace.
chap.audit.read
Read the audit log.
chap.audit.verify_chain
Verify the audit hash chain.

What it does

Single-participant calls
Every call is made as one named agent and cannot impersonate a human.
Audit chain
Reads the audit log and verifies its hash chain.
Review gating
Tasks flagged review_required cannot be completed by the agent alone.
Pluggable storage
Runs in memory, in a SQLite file, or forwards to a CHAP server.
Filtered tool sets
Choose the agent subset or every method.
HTTP handler included
The package also serves MCP over HTTP for your own server.

Good for

  • Keep a trail of which agent outputs a human approved or edited
  • Gate an assistant's task completion behind a person's review
  • Verify the integrity of an audit log
  • Add a human decision desk to an agent project

CHAP Coordinator FAQ

Can the agent approve its own work?

No. The coordinator refuses decisions on the agent's own work and logs the attempt.

Where is data kept?

In memory by default; set a database path for SQLite or point it at a CHAP server.

What runtime is needed?

Node 22 or newer.

Can it act as a human?

No. The server refuses a human participant URI at start.

How do I add CHAP Coordinator to Claude Code?

Run this command in your terminal: claude mcp add chap -- npx -y @brightbeamai/chap-coordinator-mcp

What are the best alternatives to CHAP Coordinator?

The closest alternatives to CHAP Coordinator are Chipp Agents, Two-Key, Approval Gate MCP, Chipp Agent settings and mcpcut.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers