@jgalego/teamapi-mcp-server

@jgalego/teamapi-mcp-server

An MCP server exposing a resolved Team API as Code org graph as tools for LLM assistants: list_teams, get_team, get_team_roles, get_team_cognitive_load, find_service_owner, list_services, get_team_interactions, get_team_dependencies, get_context_map, render_org_diagram, search_org, get_org_graph, get_org_cognitive_load_report, and get_org_gaps (the accountability holes between teams).

Each AI-native document domain adds a list_*/get_* pair — list_agents/get_agent, list_prompts/get_prompt, and so on — alongside render_prompt, get_context_bundle, get_knowledge_graph and traverse_knowledge_graph.

Normally started via teamapi serve-mcp — point Claude Desktop or Claude Code at that command.

Install

npm install @jgalego/teamapi-mcp-server

Usage

import { OrgGraphStore } from "@jgalego/teamapi-core";
import { createMcpServer } from "@jgalego/teamapi-mcp-server";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const store = new OrgGraphStore({ seedUris: [...] });
await store.load();

const server = createMcpServer(store);
await server.connect(new StdioServerTransport());

Claude Desktop / Claude Code

Add an entry to Claude Desktop's claude_desktop_config.json (or Claude Code's MCP config):

{
  "mcpServers": {
    "teamapi": {
      "command": "teamapi",
      "args": ["serve-mcp", "/absolute/path/to/your/org"]
    }
  }
}

Use an absolute path for both command and the org directory/pattern argument — Desktop spawns this as a subprocess without your shell's PATH, so a bare teamapi only resolves if it's on the system-wide PATH (e.g. installed via npm install -g @jgalego/teamapi); otherwise point command at the full path to the installed binary (e.g. from which teamapi).

Full docs and examples: https://github.com/JGalego/TeamAPI

License

MIT