Ateam MCP

What ariekogan/ateam-mcp MCP server does

ariekogan/ateam-mcp MCP server connects an MCP client to the ADAS platform for building and operating multi-agent solutions. It exposes tools for learning the platform specification, reviewing examples and workflows, designing skills, editing solution definitions, deploying changes, and checking runtime state.

The server supports both development and operational tasks. An agent can create or modify skills, add connectors and plugins, validate definitions before deployment, run skill or voice tests, inspect asynchronous execution chains, and list or inspect deployed solutions. Destructive operations such as deleting a solution, skill, or connector require explicit confirmation.

How it works

The MCP process communicates with an AI environment through stdio or HTTP, then sends authenticated HTTPS requests to the ADAS external API and Core runtime. The onboarding sequence starts with ateam_bootstrap, which returns platform guidance and example solution patterns. ateam_get_spec, ateam_get_examples, and ateam_get_workflows provide the reference material needed to author compatible definitions.

Tenant-aware operations require an explicit ateam_auth call. An API key can be obtained from the A-Team API-key page. Environment configuration alone does not replace this tool call. Global specification, example, and validation operations can work without tenant authentication.

Deployment targets the main branch when using ateam_build_and_run. A typical GitHub-backed workflow edits dev, reviews the diff, promotes changes to main, and then deploys. For smaller changes, ateam_patch can update selected fields and redeploy them. Its default array behavior merges protected arrays rather than silently replacing them, and dry_run:true can preview changes.

Setup and configuration

Install the published package with:

npx -y @ateam-ai/mcp

For local development, the repository instructions use git clone, npm install, and npm start. The MCP client configuration examples provide two environment variables:

• ADAS_TENANT: ADAS tenant identifier.
• ADAS_API_KEY: API key used for tenant access.

The server can be configured in Claude Desktop, Claude Code, Cursor, Windsurf, or another MCP-compatible environment. The material also describes a hosted HTTP endpoint at https://mcp.ateam-ai.com.

Tools and capabilities

The provided tool set covers:

• Bootstrapping, authentication, specification lookup, semantic specification search, examples, workflows, and design advice.
• Creating, patching, inspecting, listing, and deleting solutions, skills, and connectors.
• Deploying the main branch, uploading connector code, scaffolding connectors and UI plugins, and inspecting widget catalogs.
• Testing skills, pipelines, voice flows, notifications, conversations, connectors, and asynchronous jobs.
• Polling chain status, retrieving execution trees, and aborting running skill tests.

ariekogan/ateam-mcp MCP server is suited to conversational solution authoring, iterative deployment, integration testing, and runtime inspection. It is less suitable if you only need a generic MCP registry, a standalone model runtime, or a local tool unrelated to ADAS.

Limitations and notes

Deployment and tenant-specific inspection require authentication and access to the ADAS platform. ateam_conversation is asynchronous and returns a chain identifier; callers should poll ateam_chain_status until the complete chain finishes rather than repeatedly fetching the full chain tree. ateam_build_and_run is a heavy operation that can take 60–180 seconds and always deploys main.

Deletion operations are irreversible within ADAS Core and require confirmation parameters. Connector deletion can break skills that reference its tools. The README lists a smaller, older tool set using adas_* names, while the supplied current tool inventory uses ateam_* names; integrations should follow the current inventory exposed by the server.

The project is released under the MIT license.