Quickstart with MCP
Connect RubberDuck to your IDE and index your first repository in a few minutes.
Connect your IDE
Paste the setup prompt into Cursor, Claude Code, or Codex, or install the RubberDuck extension in VS Code, to connect the MCP servers.
Restart and verify
Confirm both MCP servers are configured and reachable.
Connect your code
Index a GitHub repository or a local project.
Ask your first question
Try a real prompt and see a grounded, evidence-backed answer.
Prerequisites
- A RubberDuck account (sign-up takes a minute)
- Cursor, Claude Code, or Codex installed, or VS Code with an AI assistant
- A repository to index, on GitHub or on your machine
New here? The wizard starts after sign-up.
Sign Up & Get StartedStep 1: Copy the setup prompt
Cloud agents don't read mcp.json, so the prompt and config below can't reach them, and a .cursor/mcp.json in your repository would only commit your token. Cursor offers cloud agents on paid plans only. Add RubberDuck in your Cursor account instead:
https://codebase.rubberduck.com/mcp
Bearer YOUR_TOKEN
https://semantic.rubberduck.com/mcp
Add these for yourself, not your whole team. A team-wide entry carries your token, so every teammate's agent connects as you: one RubberDuck user, with your repositories and your usage.
Paste this into Cursor Chat. Replace YOUR_TOKEN with the token from your dashboard.
Please configure and connect me to these MCP servers.
First check where you are running. If you are a Cursor cloud agent (started from cursor.com/agents, Slack, GitHub or the web, in a remote VM rather than on my computer), don't create or edit any mcp.json: cloud agents ignore mcp.json, and a file in the repository would commit my token. Instead, stop and give me these steps to do myself:
1. Open https://cursor.com/agents, open its MCP settings and choose Add MCP, then Custom MCP.
2. Name it rubberduck-codebase-intelligence, set Server URL to https://codebase.rubberduck.com/mcp, and under Headers add the key Authorization with the value "Bearer <the token in the config below>". Paste the token itself; don't use ${env:...}, cloud agents send it unchanged. Click Add MCP.
3. Add rubberduck-semantic-intelligence the same way, with the Server URL https://semantic.rubberduck.com/mcp and the same Authorization header.
4. Start a new cloud agent. Agents that were already running keep the MCP servers they started with.
Otherwise, in Cursor on my computer, add this to my user-level Cursor MCP config at ~/.cursor/mcp.json (not .cursor/mcp.json inside the project, which would put my token in the repository). If the file already has an "mcpServers" section, add these two entries to it:
{
"mcpServers": {
"rubberduck-codebase-intelligence": {
"url": "https://codebase.rubberduck.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
},
"rubberduck-semantic-intelligence": {
"url": "https://semantic.rubberduck.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN"
}
}
}
}
After saving the config, ask me to restart the session so I can connect to the MCPs.The prompt failed, or your IDE showed an error? Open MCP Troubleshooting for the fix.
Step 2: Restart and verify
After restarting your IDE, paste this to confirm both servers are connected:
Verify that both MCP servers are properly configured and reachable. Respond only with "healthy" or "unhealthy." If they are healthy, use the Rubberduck MCP server wherever applicable in subsequent tasks and prompts.
Got an error, or answered “unhealthy”? Open MCP Troubleshooting for the fix.
Step 3: Connect your code
- Install the RubberDuck GitHub App and select your repos
- Paste this into your IDE (replace with your repo URL):
Index my GitHub repo: https://github.com/your-org/your-repo