Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.armature.tech/llms.txt

Use this file to discover all available pages before exploring further.

Armature exposes an agent-facing MCP endpoint at /api/mcp. Any MCP-compatible client—Claude Code, Claude Desktop, Cursor, Windsurf, or your own tooling—can connect to it using an Armature API key. Once connected, the agent has access to triage tools for investigating run failures, repair and proposal tools for patching workflows, run dispatch tools for triggering validations on demand, structured resources for traces and coverage reports, and guided prompts for common workflows. All tool responses include both a text JSON representation and a structuredContent block, so agents can work with results programmatically without additional parsing.

​
Endpoint URL

Your Armature MCP endpoint is your organization’s dashboard URL with /api/mcp appended: 
https://your-org.armature.app/api/mcp

​
Connecting an MCP client

To connect a client, add your Armature endpoint and API key to the client’s MCP server configuration. The key goes in the Authorization header as a Bearer token—see Authentication for how to generate one. 
{
  "mcpServers": {
    "armature": {
      "url": "https://your-org.armature.app/api/mcp",
      "headers": {
        "Authorization": "Bearer amt_<key-id>_<secret>"
      }
    }
  }
}
Replace your-org with your Armature organization slug and amt_<key-id>_<secret> with a real API key. The key’s role determines which tools and operations are available to the agent—see Roles for the full permission matrix.

​
What the endpoint exposes

The MCP endpoint is organized into five surfaces: Triage tools let the agent search, inspect, and diagnose workflow runs, and retrieve overall organization health. These are read-only and safe to call at any time. Repair and proposal tools let the agent stage a patch against a workflow, draft a regression workflow from a failed run, and apply an approved change. Applying a change requires editor role or higher. Run tools let the agent dispatch a manual workflow run with an optional wait for completion, and compare two runs to surface regressions. Resources expose structured data—run traces, run artifacts, workflow schemas, the model catalog, evaluation criteria documentation, MCP server tool catalogs, coverage reports, and staged change proposals—as addressable armature:// URIs. Prompts provide guided multi-step sequences for the most common agent tasks: repairing a failing workflow and designing a workflow test suite.

​
Available surfaces at a glance

Authentication

Generate API keys, set the Authorization header, and understand token security requirements.

Roles

Understand which tools and operations each role can access through the MCP endpoint.

Triage tools

Search runs, inspect run details, diagnose failures, and check organization health.

Repair tools

Propose workflow patches, draft regression workflows, and apply approved changes.

Resources

Access run traces, coverage reports, workflow schemas, and staged proposals by URI.

Prompts

Use guided multi-step prompts to repair failing workflows and design test suites.