Claude Cowork Integration Guide

Last Updated: January 22, 2026 (Session 61)

Overview

This guide explains how to use GoGrab MCP from Claude Cowork (claude.ai/cowork) to give Claude access to your AI intelligence layer - past decisions, learned patterns, and specialized agent experts.

With MCP integration, Claude can:

• Remember decisions across conversations
• Learn from successful patterns
• Route tasks to specialized agent experts
• Provide context-aware suggestions

Quick Start

Step 1: Get Your API Credentials

1. Visit gograb.space
2. Click "Get Started"
3. Enter your email to receive a magic link
4. Click the link to access your MCP Setup page
5. Copy your API Key and Project ID

Step 2: Store Your Credentials

Save these as environment variables or store securely:

MCP_API_KEY="mcp_your_api_key_here"
MCP_PROJECT_ID="your-project-id"
MCP_BASE_URL="https://ai-api.gograb.space/ai-api/mcp"

Step 3: Use MCP in Conversations

In Claude Cowork, you can reference MCP by describing what you want. Claude can make HTTP calls to the MCP API.

Using MCP with Claude Cowork

Before Starting Any Task: Call Preflight

Tell Claude to get context before starting significant work:

Before we begin, please call the GoGrab MCP preflight endpoint:

POST https://ai-api.gograb.space/ai-api/mcp/preflight
Headers:
  X-MCP-API-Key: mcp_your_api_key
  X-MCP-Project-Id: your-project-id
  Content-Type: application/json
Body: {"task": "Implementing user authentication with JWT"}

Claude will receive:

• Expert routing - Which specialized agent is best for this task
• Historical hints - Past decisions relevant to the task
• Patterns - Learned patterns that worked before
• Conventions - Codebase conventions to follow

After Making Important Decisions: Log Them

When Claude helps you make a significant decision, ask it to log the decision:

Great, please log this decision to MCP:

POST https://ai-api.gograb.space/ai-api/mcp/tools/call
Headers:
  X-MCP-API-Key: mcp_your_api_key
  X-MCP-Project-Id: your-project-id
  Content-Type: application/json
Body:
{
  "name": "log_decision",
  "arguments": {
    "type": "implementation",
    "question": "Which authentication strategy to use?",
    "chosen": "JWT with refresh tokens",
    "reasoning": "Stateless, scalable, works with mobile apps",
    "confidence": 0.9
  }
}

Finding Past Decisions

Ask Claude to search for similar past decisions:

Can you search MCP for past decisions about "database indexing"?

POST https://ai-api.gograb.space/ai-api/mcp/tools/call
Headers:
  X-MCP-API-Key: mcp_your_api_key
  X-MCP-Project-Id: your-project-id
  Content-Type: application/json
Body:
{
  "name": "find_similar",
  "arguments": {
    "query": "database indexing",
    "limit": 5
  }
}

API Reference

Authentication

All requests require these headers:

Endpoints

Preflight - Get Context Before Task

POST /ai-api/mcp/preflight

{
  "task": "Brief description of your task"
}

Response:

{
  "success": true,
  "preflight": {
    "task": "Brief description...",
    "guidance": "🎯 Expert: Backend Agent. Preflight for project:my-project",
    "expert": {
      "agent": "backend",
      "name": "Backend Agent",
      "confidence": 0.85,
      "capabilities": ["api-design", "data-modeling", "authentication"]
    },
    "historicalHints": [
      {"type": "implementation", "question": "...", "chosen": "..."}
    ]
  }
}

Call Tool - Execute Any MCP Tool

POST /ai-api/mcp/tools/call

{
  "name": "tool_name",
  "arguments": { ... }
}

List Tools - See Available Tools

GET /ai-api/mcp/tools

Available Tools

Agent Experts

MCP routes your tasks to specialized agent experts:

The preflight response includes which expert was selected and why.

Example Session

Here's a complete example of using MCP in Claude Cowork:

1. Start with Preflight

You: "I need to implement a rate limiter for our API. Can you first check MCP for any relevant past decisions?"

Claude: "I'll call the MCP preflight endpoint to get context."

curl -X POST https://ai-api.gograb.space/ai-api/mcp/preflight \
  -H "X-MCP-API-Key: mcp_abc123" \
  -H "X-MCP-Project-Id: my-project" \
  -H "Content-Type: application/json" \
  -d '{"task": "Implementing API rate limiter"}'

Response: "The Backend Agent expert was selected with 85% confidence. Found 3 relevant past decisions about rate limiting..."

2. Make a Decision

You: "Based on that context, let's use Redis with sliding window. Please log this decision."

Claude: "I'll log this decision to MCP."

curl -X POST https://ai-api.gograb.space/ai-api/mcp/tools/call \
  -H "X-MCP-API-Key: mcp_abc123" \
  -H "X-MCP-Project-Id: my-project" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "log_decision",
    "arguments": {
      "type": "architecture",
      "question": "Which rate limiting strategy for API?",
      "chosen": "Redis sliding window rate limiter",
      "reasoning": "Distributed, handles multiple instances, proven pattern from past decisions",
      "confidence": 0.9
    }
  }'

3. Search for Similar Decisions Later

You: "What did we decide about rate limiting before?"

Claude: "Let me search MCP."

curl -X POST https://ai-api.gograb.space/ai-api/mcp/tools/call \
  -H "X-MCP-API-Key: mcp_abc123" \
  -H "X-MCP-Project-Id: my-project" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "find_similar",
    "arguments": {
      "query": "rate limiting API",
      "limit": 5
    }
  }'

Tips for Best Results

1. Always start with preflight - It gives Claude important context
2. Log significant decisions - This builds your knowledge base
3. Be specific in task descriptions - Better routing to the right expert
4. Report outcomes - Use report_outcome to tell MCP if decisions worked
5. Search before deciding - Use find_similar to learn from past decisions

OpenAPI Documentation

Full API documentation is available at:

• OpenAPI Spec: https://ai-api.gograb.space/ai-api/mcp/openapi.yaml
• OpenAI Functions: GET /ai-api/mcp/openai/functions

Troubleshooting

"API key authentication requires both headers"

Make sure you include both:

• X-MCP-API-Key
• X-MCP-Project-Id

"Project not found"

Verify your project ID is correct. List your projects with JWT auth:

curl https://ai-api.gograb.space/ai-api/mcp/projects \
  -H "Authorization: Bearer YOUR_JWT"

"Hard limit reached"

You've exceeded your monthly API call limit. Check usage:

curl https://ai-api.gograb.space/ai-api/mcp/usage \
  -H "X-MCP-API-Key: mcp_abc123" \
  -H "X-MCP-Project-Id: my-project"

Related Documentation

• Cloud MCP External Access
• MCP Superpower Vision
• OpenAPI Specification

Session 61 - Claude Cowork Integration Guide