Debug Incident MCP Server

An MCP (Model Context Protocol) server for uploading debug incidents to Debug Memory via Cursor.

Installation

1. Install the MCP Server

Clone or download this repository:

git clone https://github.com/yourusername/debug-mcp.git
cd debug-mcp

2. Install Dependencies

Using uv (recommended):

uv sync

This creates a virtual environment at .venv and installs all dependencies.

Or install from PyPI using uv:

uv pip install debug-mcp

3. Get Your API Key

1. Go to your Debug Memory dashboard
2. Navigate to Settings → API Keys
3. Create a new API key
4. Copy the key (starts with dm_)

4. Configure Cursor

See Cursor Setup Guide for detailed instructions.

Quick config - create or edit .cursor/mcp.json in your project:

{
  "mcpServers": {
    "debug-incident": {
      "command": "uvx",
      "args": ["debug-mcp"],
      "env": {
        "WORKSPACE_API_KEY": "dm_your_api_key_here"
      }
    }
  }
}

⚠️ Replace:

• dm_your_api_key_here with your API key from Debug Memory

5. Restart Cursor

Close and reopen Cursor for the MCP server to load.

Usage

Once configured, the MCP server exposes these tools (see docs/tools.md for the full workflow and next_action behavior):

1) ranked_solutions

• Search similar incidents and return env-aware ranked solutions
• Inputs: query_text (required), env (required), limit (optional, default 5)
• Output: lookup_id, ranked solutions, recommended solution, next_action

2) add_incident

• Create a new incident + solution + outcome (blocks duplicates if a similar incident already exists)
• Inputs: title, summary, error_signature, tags, steps, env, worked, notes (optional)

3) add_solution

• Add a new fix (or env-specific variant) for an existing incident
• Inputs: incident_id (required), steps (required), env (required), env_bucket (optional)
• Next action: Record an outcome for this solution

4) record_outcome

• Record whether a solution worked or failed
• Inputs: solution_id (required), worked (required), env (required), env_bucket (optional), notes (optional)

Examples

Upload an incident:

Upload an incident: "TypeError in auth module"

Upload with details:

Upload an incident with title "API timeout in production" 
with summary "Payment API timing out after 30s" 
and tags ["api", "performance", "production"]

Search for similar incidents:

Search for incidents about "null pointer exception"

Find similar incidents to "cannot read property of undefined"

Workflow example:

1. "Search for incidents about TypeError in map function"
2. (Review results, if no solution found)
3. "Upload incident: TypeError when calling map on undefined array, 
    fixed by adding null check, root cause was API returning null"

How It Works

1. The MCP client calls tools like ranked_solutions, add_solution, record_outcome, add_incident
2. MCP server calls Debug Memory edge function endpoints
3. Edge function authenticates your API key and creates the incident
4. MCP server generates embeddings using all-MiniLM-L6-v2 model
5. Embeddings are uploaded for title, summary, and error_signature fields
6. Incident is stored in your workspace with semantic search enabled

Architecture

┌─────────┐           ┌─────────────┐           ┌──────────────┐
│ Cursor  │  ◄─MCP─►  │ MCP Server  │  ◄─HTTP─► │ Edge Function│
│   AI    │   stdio   │  (Python)   │   POST    │  (Supabase)  │
└─────────┘           └─────────────┘           └──────────────┘
                             │                          │
                             │                          ▼
                      Configured with            ┌──────────────┐
                      WORKSPACE_API_KEY          │   Database   │
                                                 └──────────────┘

Features

• 🚀 Easy Integration: One-command setup with uv
• 🔒 Secure: API keys stay in config, never exposed to LLM
• 🧠 Semantic Search: Auto-generates embeddings for intelligent search
• 🔍 Find Solutions: Search past incidents to find fixes for similar problems
• 📝 Rich Incident Data: Capture title, summary, error signatures, fixes, and more
• 🏷️ Tagging: Categorize incidents with custom tags

Security

• ✅ API key stored in Cursor config (not in code)
• ✅ API key never exposed to LLM or conversation logs
• ✅ Edge function validates API key server-side
• ✅ No direct database access from MCP server
• ✅ Embeddings generated locally before upload

Technical Details

• Embedding Model: all-MiniLM-L6-v2 (384 dimensions)
• Transport: stdio (local execution)
• Dependencies: Managed with uv
• Python: 3.10+

Files