The MCP server for Cursor that serves specs, schemas, and backlog.

Cursor is one of the fastest AI editors, but it suffers from a fundamental context dilemma: prompts and .cursorrules are static, unversioned, and consume precious context window.

When you paste requirements into Cursor Composer, you are feeding unstructured prose. Cursor guesses the foreign keys in your database, guesses whether a user should be redirected, and tests buttons instead of underlying business logic.

ContextsBase solves this by acting as the context infrastructure for Cursor. Instead of dumping your app’s entire history into prompt tokens, Cursor calls ContextsBase over MCP on demand. It receives:

• Structured Entities: JSON data models with fields, types, and relations so Cursor never invents a column.
• Business Rules (F-n): Functional specs with explicit acceptance criteria and edge cases.
• Given / When / Then Tests (T-n): Rules Cursor automatically converts into Playwright assertions.
• Atomic Claims: Iteration items claimed one by one so multi-turn sessions don't duplicate work.

You can configure ContextsBase in Cursor in two ways: via local stdio process (npx) or direct HTTP SSE connection.

Option A: Workspace .cursor/mcp.json (Recommended)

Create a file at .cursor/mcp.json in your project repository root. Paste this snippet:

{
  "mcpServers": {
    "contextsbase": {
      "command": "npx",
      "args": ["-y", "@contextsbase/mcp-server"],
      "env": {
        "CONTEXTSBASE_TOKEN": "cb_tok_live_abc123..."
      }
    }
  }
}

Option B: Direct Remote HTTP SSE

If your team prefers not to run local Node processes or wants Cursor connected across remote VMs:

{
  "mcpServers": {
    "contextsbase": {
      "url": "https://app.contextsbase.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer cb_tok_live_abc123..."
      }
    }
  }
}

Verify Connection in Cursor

Open Cursor Settings > Features > MCP. You will see contextsbase with a green status indicator and 5 tools enabled.

ContextsBase provides five purpose-built MCP tools. Cursor calls them automatically during Composer and Agent sessions:

Once connected, you no longer need to write multi-paragraph prompts describing your database schema or acceptance criteria. Simply trigger Cursor Composer (Cmd + I or Ctrl + I) with:

Check ContextsBase next_iteration_item, implement the claimed feature, and run the Playwright test cases.

Here is what happens under the hood:

1. 01
   Atomic Claim: Cursor invokes next_iteration_item. ContextsBase locks the next pending feature (e.g. F-4) to prevent collisions.
2. 02
   Schema & Rules Ingestion: The payload delivers exact TypeScript/SQL entity definitions and Given/When/Then test cases into Cursor’s session.
3. 03
   Code & Playwright Generation: Cursor writes the components and produces a matching Playwright spec.
4. 04
   Feedback & Recording: Cursor calls record_test_automation and record_feature_implementation. The feature moves to Implemented with a cryptographic fingerprint of the spec.

Watch how an agent pulls from ContextsBase over MCP, implements a spec, and updates the backlog:

Most developers start with a .cursorrules file. Here is how that scales as your codebase grows: