ateam-mcp is an MCP (Model Context Protocol) server designed to give any AI environment the ability to build, validate, and deploy production-grade multi-agent systems. It is developed by ariekogan and hosted on GitHub, and by connecting the ADAS platform directly to AI assistants, it makes AI the developer's operating interface.
Purpose
ateam-mcp solves the platform knowledge barrier and tool-switching problems in multi-agent system development. The traditional approach requires developers to have an in-depth understanding of the ADAS platform, configure it manually, and switch back and forth between documentation, editors, and dashboards. ateam-mcp exposes these capabilities to AI assistants in natural language form, letting the AI assistant handle the entire lifecycle: reading specifications, building skill definitions, validating, deploying to production, and verifying health status.
Main capabilities
- Connects AI assistants with the ADAS platform: supports ChatGPT, Claude, Gemini, Copilot, Cursor, Windsurf, and any MCP-compatible environment.
- Provides 12 ADAS tools, covering specification reading, skill definition building, validation, deployment, and monitoring.
- Supports hosted connectors: by pasting a URL (https://mcp.ateam-ai.com）即可在 ChatGPT developer mode, claude.ai, Claude Desktop, Claude Code, and other environments.
- Supports local processes: run via npx -y @ateam-ai/mcp, and log in using ateam_auth.
- Supports remote HTTP endpoints:https://mcp.ateam-ai.com（也支持 /mcp), suitable for any client that supports Streamable HTTP transport.
- Provides workspace management tools (ateam_workspace_*), available after logging in.
Usage
ChatGPT
In Settings → Connectors → Developer mode, paste https://mcp.ateam-ai.com，所有 12 ADAS tools will then appear in ChatGPT. ChatGPT Pro, Plus, Business, or Enterprise users can all build and deploy multi-agent solutions through conversation.
claude.ai and Claude Desktop
Open Customize > Connectors (in Claude Desktop, Customize in the sidebar, then Connectors), click Add custom connector, enter https://mcp.ateam-ai.com，点击 Connect. The browser will open the A-Team login page and ask for the workspace key: copy it from app.ateam-ai.com/connect (requires workspace owner or admin), then paste it. The key determines the workspace; to switch workspaces, disconnect and reconnect using the key for another workspace.
Claude Code
Use the hosted connector, log in the same way, then run /mcp and authenticate in the browser.
Cursor / Windsurf / VS Code (Copilot)
Add the hosted server to .cursor/mcp.json, mcp_config.json, or .vscode/mcp.json. Clients that support OAuth will open the A-Team login page; clients that do not support it will send the key from their own configuration as Authorization: Bearer <key>. The local process (npx -y @ateam-ai/mcp) only logs in through ateam_auth: ADAS_API_KEY in the environment will not log it in.
Gemini and other platforms
Remote HTTP endpoint (https://mcp.ateam-ai.com，也支持 /mcp) can be used with any client that supports Streamable HTTP transport. The HTTP endpoint requires credentials on every request: OAuth (supported clients will automatically follow the 401 challenge), or an A-Team API key sent as Authorization: Bearer <key>. Anonymous requests are rejected with 401; as of 0.4.93, there is no anonymous session available for calling ateam_auth.
Local process
The local process logs in using ateam_auth and requires the agent to read the workspace key from a file outside the chat; ADAS_API_KEY in the environment will not make it log in, and will never be sent. ADAS_API_URL in the environment can point it to a self-hosted API. Most people use the hosted connector.
Use cases
- Developers who want to quickly build and deploy multi-agent systems through natural language, such as customer support systems (including order tracking and escalation).
- Teams that want to reduce switching between documentation, editors, and dashboards to improve development efficiency.
- Developers using AI tools such as ChatGPT, Claude, and Cursor who want to access ADAS platform capabilities directly in their existing environment.
Limitations
- Requires a workspace key or API key for the ADAS platform to use.
- Anonymous requests are rejected (401).
- The local process logs in only through ateam_auth; the environment variable ADAS_API_KEY is invalid.
- Trial workspaces or keys not bound to a person will be rejected by the platform, and a single-file tool must be used instead.
- Specific fees, pricing, and platform support details are not provided for now.
How to obtain
- npm: npm search mcp ai-agents → @ateam-ai/mcp
- Official MCP registry: registry.modelcontextprotocol.io
- Claude Desktop extension: built-in extension browser
- Claude Code plugin marketplace: /plugin → Discover tab
- Windsurf MCP marketplace: built-in marketplace
- VS Code MCP gallery: Extensions view
- Community directories: Smithery, mcp.so, PulseMCP (30,000+ listings in total)
