Cursor+ MCP Bundles

How Cursor Uses MCP Tools

When Cursor starts, it reads your .cursor/mcp.json configuration and connects to every MCP server listed. Each server responds with its available tools — names, descriptions, and parameter schemas. Cursor's AI agent then treats these tools exactly like built-in capabilities: it can call them, process results, and chain multiple calls together within a single conversation.

This is fundamentally different from copy-pasting API responses into a chat window. Cursor's agent sees MCP tool results as structured data it can act on immediately — generating code, writing files, running terminal commands, or calling additional tools based on what it learned.

Two Ways to Set Up MCP in Cursor

Method 1: MCPBundles CLI (Recommended)

The fastest path from zero to working tools:

pip install mcpbundles
mcpbundles init

This detects your operating system, finds your Cursor config directory, discovers your enabled bundles, and writes the correct .cursor/mcp.json entries automatically. No manual JSON editing, no copying URLs.

After init completes, restart Cursor. Your tools are ready.

For power users, the CLI also provides:

• mcpbundles tools — browse all available tools across your bundles
• mcpbundles call <tool-name> — test any tool directly from your terminal
• mcpbundles shell — interactive REPL for exploring tools

Method 2: Manual mcp.json Configuration

Create or edit .cursor/mcp.json (global at ~/.cursor/mcp.json, or per-project at .cursor/mcp.json in your project root):

{
  "mcpServers": {
    "mcpbundles": {
      "url": "https://mcp.mcpbundles.com/bundle/your-bundle-slug"
    }
  }
}

Replace your-bundle-slug with the slug from your MCPBundles dashboard. You can add multiple servers:

{
  "mcpServers": {
    "marketing": {
      "url": "https://mcp.mcpbundles.com/bundle/marketing-tools"
    },
    "dev-ops": {
      "url": "https://mcp.mcpbundles.com/bundle/devops-suite"
    }
  }
}

Cursor connects to all servers on startup and makes every tool available to the AI agent.

What Cursor Can Do With MCP Tools

Code Generation From Live Data

Ask Cursor to fetch real data from your APIs and generate code that matches the actual response shape. Instead of hallucinating a type definition, Cursor calls your HubSpot bundle, inspects the contact schema, and writes an accurate TypeScript interface.

Cross-Service Workflows

Cursor's agent can chain tool calls across services. For example: "Pull the latest Stripe invoices, find overdue ones, look up the customers in HubSpot, and draft follow-up emails in Gmail." Each step uses a different MCP tool, and Cursor orchestrates the entire flow.

Database Queries and Migrations

Connect a PostgreSQL or MySQL bundle and ask Cursor to query your database, analyze the schema, generate migrations, or write seed data — all without leaving the editor.

DevOps and CI/CD

With GitHub, GitLab, or Jira bundles connected, Cursor can review pull requests, create issues, check CI status, and correlate deployment logs with code changes.

Real-Time Debugging

When debugging an issue, Cursor can call monitoring tools (Datadog, PostHog, Sentry) to fetch recent errors, correlate them with your code, and suggest fixes — all in one conversation.

Agent Mode, Ask Mode, and Background Agents

Cursor offers three interaction modes, and MCP tools work in all of them:

Agent Mode gives the AI full autonomy to read files, run commands, and call MCP tools. This is the most powerful mode for multi-step workflows — the agent decides which tools to call and how to use the results.

Ask Mode is read-only by default but still calls MCP tools when you ask. Use this when you want data from external services without giving the AI write access to your codebase.

Background Agents run autonomously on long tasks. They can call MCP tools as part of multi-step plans — for example, fetching data from an API, generating a report, and committing it to a branch.

Security and Credentials

MCPBundles handles all credential management. When Cursor calls an MCP tool, the request goes to your MCPBundles server, which injects the correct API key or OAuth token before forwarding to the provider. Cursor never sees raw credentials.

All credentials are encrypted at rest using Fernet symmetric encryption. OAuth tokens refresh automatically. You can revoke access to any provider at any time from the MCPBundles dashboard, and the change takes effect immediately — no Cursor restart needed.

Tips for Getting the Most Out of Cursor + MCP

Be specific about what you want. Instead of "get my data," say "fetch the 10 most recent contacts from HubSpot with their email, company, and last activity date." Specific prompts produce better tool calls.

Use project-level configs. Put a .cursor/mcp.json in each project root with only the bundles relevant to that project. A marketing project gets CRM and email tools; a backend project gets database and DevOps tools.

Combine MCP tools with codebase context. Cursor's strength is understanding your code. Ask it to "update the Stripe webhook handler to match the current event schema" and it will call the Stripe bundle to get the schema, then modify your code accordingly.

Chain multiple tools in one prompt. Cursor's agent handles multi-step workflows naturally. "Check our GitHub PRs, find any that touch the billing module, cross-reference with recent Stripe disputes, and create a summary" works as a single request.