Quick Start

Two minutes. That’s all it takes to give your AI access to 100,000+ API operations.

1. Add the MCP Server

  1. Open Settings → Connectors
  2. Click Add custom connector
  3. Enter:
    • Name: Toolcog
    • URL: https://mcp.toolcog.com
  4. Click Add

Claude will prompt you to sign in when it connects.

For other clients, see AI Clients.

2. Start a Conversation

Open a new conversation and try something:

“What GitHub operations are available for working with issues?”

Claude will use Toolcog to discover relevant operations—github/createIssue, github/updateIssue, github/listIssues, and more. No configuration. No API keys yet. Just ask.

Try another:

“Show me what parameters I need to create a Stripe customer.”

Claude discovers the operation, retrieves the interface, and shows you exactly what’s required. This works because discovery and learning don’t require authentication—AI is just reading API specifications.

3. Make It Real

Now ask Claude to actually do something:

“Create a GitHub issue in my repo called ‘Test from Toolcog’”

This time, Claude needs your GitHub credentials. Instead of failing, it responds with something like:

“I need authorization to access GitHub. Click here to connect: [authorization link]”

Click the link. You’ll be taken through GitHub’s OAuth flow. Grant permission, and you’re redirected back. Now try again—the issue gets created.

That’s it. You just gave AI the ability to discover, learn, and execute any API operation, with authentication that happens seamlessly when needed.

What Just Happened

Behind the scenes, Claude used three API meta-tools:

  1. find_api — Searched 100,000+ operations to find what you needed
  2. learn_api — Retrieved the TypeScript interface for the operation
  3. call_api — Executed the operation with your credentials

You didn’t configure which APIs to use. You didn’t specify which operations were available. You asked Claude to do something, and the system figured out the rest.

Next Steps