Claude Desktop Setup
Connect Claude Desktop to ACE to access your playbooks directly in conversations.
Prerequisites
- Claude Desktop installed (macOS or Windows)
- ACE account with verified email
- API key with
playbooks:readscope (minimum)
Configuration
Step 1: Create an API Key
- Go to app.aceagent.io
- Navigate to API Keys
- Create a key with all scopes enabled:
playbooks:read- List and read playbooksplaybooks:write- Create playbooks and versionsoutcomes:write- Record task outcomesevolution:write- Trigger evolutionevolution:read- Check evolution status
Step 2: Open Config File
- Open Claude Desktop
- Click the Settings icon
- Select the Developer tab
- Click Edit Config
This opens claude_desktop_config.json in your default editor.
Step 3: Add MCP Server
Add the ACE MCP server to the config file:
{
"mcpServers": {
"ace": {
"type": "sse",
"url": "https://aceagent.io/mcp/sse",
"headers": {
"X-API-Key": "YOUR_API_KEY"
}
}
}
}
Replace YOUR_API_KEY with your actual API key.
Step 4: Restart Claude Desktop
- Quit Claude Desktop completely
- Reopen the application
- The ACE tools should now be available
Verifying Connection
Check Available Tools
In Claude Desktop, ask:
"What ACE tools are available?"
Claude should list the available tools:
list_playbooks/find_playbookget_playbookcreate_playbook/create_versionrecord_outcometrigger_evolution/get_evolution_status
Test a Tool
Try listing your playbooks:
"List my ACE playbooks"
Using ACE in Claude Desktop
Retrieve a Playbook
"Get my code review playbook using ACE"
Claude will use get_playbook and display the content.
Follow Playbook Instructions
"Using my code review playbook, review this code: [paste code]"
Claude will fetch the playbook, follow the instructions, and provide a review.
Record an Outcome
After a task:
"Record an outcome for the code review playbook. It was successful and caught a security issue."
Claude will use record_outcome with your feedback.
Workflow Examples
Code Review Workflow
You: "Get my TypeScript code review playbook"
Claude: [Fetches playbook content]
You: "Review this PR: [paste diff]"
Claude: [Follows playbook, provides review]
You: "Record this as a successful outcome - the review caught an injection bug."
Claude: [Records outcome with your notes]
Documentation Workflow
You: "Get my API documentation playbook"
Claude: [Fetches playbook]
You: "Document this function: [paste code]"
Claude: [Creates documentation following playbook]
You: "Record this as partial success - good structure but missed error cases"
Claude: [Records partial outcome]
Tips
Provide Detailed Outcomes
When recording outcomes, include specifics so evolution has good data to work with:
- "Record a successful outcome. The review caught 3 bugs and provided clear improvement suggestions."
- "Record this as partial - the structure was good but it missed edge cases in the error handling."
Vague feedback like "it went well" won't help evolution improve the playbook.
Use Playbook Sections
Request specific sections for focused tasks:
"Get just the Security section from my code review playbook"
Troubleshooting
Tools Not Appearing
- Check config file - Open via Settings > Developer > Edit Config
- Validate JSON syntax - Use a JSON validator
- Verify URL is correct -
https://aceagent.io/mcp/sse - Restart Claude Desktop - Full quit, not just close window
Authentication Errors
- "Unauthorized" - Check API key is correct
- "Forbidden" - Verify key has required scopes
- "Invalid token" - Verify API key format is correct
Slow Response
- First MCP request may take 2-3 seconds to establish the connection
- Subsequent requests should be faster
Security Considerations
Use Minimum Scopes
For read-only use, only grant:
playbooks:read
For full workflow:
playbooks:readoutcomes:write
Rotate Keys Regularly
- Create new key in dashboard
- Update config file
- Restart Claude Desktop
- Delete old key