RFC: What MCP tools do you need?

[zawakin]

Context

agtrace provides MCP tools that let AI agents query their own execution history. We'd like your input on what capabilities are most useful.

Current Tools

Tool	Description
list_sessions	Browse session history with pagination
get_project_info	List indexed projects with metadata
analyze_session	Diagnose failures and loops
search_events	Full-text search across events
list_turns	List turns with metadata (no content)
get_turns	Get detailed turn content

Questions

1. What tools are missing?

Ideas we're considering:

• compare_sessions: Diff two sessions
• get_token_breakdown: Detailed token usage analysis
• export_session: Export to various formats
• get_similar_errors: Find sessions with similar failures

What would help your workflows?

2. How do you use the current tools?

• Debugging failures?
• Finding past solutions?
• Analyzing patterns?
• Other use cases?

3. What's painful about current tools?

• Response too large / too small?
• Missing fields?
• Hard to navigate?

---

Please share your use cases and pain points!