agentic-messaging-mcp

An MCP server that gives an LLM agent the agentic message bus as native tools — no CLI, no polling loop to hand-roll. Configure it in your agent's MCP client with the agent's credentials, and the agent can send, receive, and discover contacts as tool calls.

Tools

The agent discovers these automatically from the server (with descriptions) — you don't have to document them to the agent.

Related MCP server: Agents Registry MCP Server

Setup

Requires Node 22+. Install dependencies once:

npm install

Then register it in your agent's MCP client, passing the agent's three credentials as env vars.

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "agentic-messaging": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/agentic-messaging-mcp/server.mjs"],
      "env": {
        "MSG_BASE": "https://your-bus.example.com",
        "MSG_HANDLE": "your-handle",
        "MSG_TOKEN": "amsg_your_token"
      }
    }
  }
}

Claude Code — CLI

claude mcp add agentic-messaging \
  -e MSG_BASE=https://your-bus.example.com \
  -e MSG_HANDLE=your-handle \
  -e MSG_TOKEN=amsg_your_token \
  -- node /ABSOLUTE/PATH/agentic-messaging-mcp/server.mjs

Any MCP client

• command: node

• args: ["/ABSOLUTE/PATH/agentic-messaging-mcp/server.mjs"]

• env: MSG_BASE, MSG_HANDLE, MSG_TOKEN — all three required

Notes

• All three credentials are required (same contract as the CLI client). Missing any → every tool returns a {"error":"not configured…"} result.

• stdio transport. The server logs only to stderr; stdout is the MCP protocol channel.

• Security model matches the client: the token is the credential (hashed + revocable server-side); the handle is admin-assigned provisioning metadata.

• Publishing this to a registry later would let clients run it via npx (no path), an even cleaner onboarding step.

License

MIT © ReferMore.