Connect your coding agent

Give Claude Code, Cursor, or any MCP client live access to your project's requirements, tasks, and tests.

Ragnarock exposes your project's context — requirements, features, tasks, and test cases — to coding agents through the Model Context Protocol (MCP). Once connected, an agent like Claude Code or Cursor can read your SRS before writing code, check test cases, and mark tasks complete as it works, all without you copying and pasting context.

How it works

You create an API key for the project (see Settings & API keys).
You paste that key into your coding agent's MCP configuration.
The agent connects to the Ragnarock MCP server and can call its tools — scoped to just that one project.

The key is how the server knows which project to serve, so each key only ever exposes the project it was created for.

Step 1 — Create a project API key

Open Settings → MCP API keys in your project, create a key, and copy it. It's shown only once. Full details in Settings & API keys.

Step 2 — Add it to your agent

Use the MCP server URL provided by your Ragnarock administrator (the default local address is http://localhost:3002/mcp). Authenticate with your key using the x-ragnarock-key header.

Add this to .claude/settings.json in your coding project:

{
  "mcpServers": {
    "ragnarock": {
      "type": "http",
      "url": "http://localhost:3002/mcp",
      "headers": {
        "x-ragnarock-key": "ragnarock_mcp_<your-key>"
      }
    }
  }
}

Claude Code discovers the tools automatically and can call them while it works.

Add this to .cursor/mcp.json:

{
  "mcpServers": {
    "ragnarock": {
      "url": "http://localhost:3002/mcp",
      "headers": {
        "x-ragnarock-key": "ragnarock_mcp_<your-key>"
      }
    }
  }
}

To explore the tools interactively without a coding agent, use the official MCP Inspector:

npx @modelcontextprotocol/inspector http://localhost:3002/mcp

Add your key under Headers → x-ragnarock-key, then browse and call the tools.

The tools your agent gets

Once connected, the agent can call these tools:

Tool	What it returns / does
`get_srs`	Your full Software Requirements Specification.
`get_features`	Product features with their requirements and counts.
`get_next_task`	The highest-priority pending task.
`get_task_detail`	Full detail for a specific task.
`mark_task_complete`	Marks a task done and returns the next one.
`get_test_cases`	Test cases for a feature.
`get_architecture_docs`	Your generated architecture documents (SAD / HLD / LLD).

A typical coding loop

With the tools connected, a coding agent can work through your backlog on its own:

Call get_next_task to pick up the highest-priority work.
Call get_srs and get_architecture_docs to understand what's required.
Write the code.
Call get_test_cases to check the acceptance criteria.
Call mark_task_complete to close the task and fetch the next one.

Managing access

Each key is scoped to a single project. Create separate keys for separate projects or devices.
Revoke a key from Settings → MCP API keys whenever a device is lost or someone leaves — the agent loses access immediately.

The MCP server only exposes data your project already produced in Ragnarock. Build a solid SRS, plan, and test suite first, and your coding agent inherits all of it.