leeroo-ai Knowledge Base -- Leeroopedia MCP Tools
You have access to Leeroopedia, a curated ML/AI knowledge base, via MCP tools. These are real MCP tools registered in your environment -- call them directly like any other tool. They contain framework-specific docs, code examples, API references, and best practices.
IMPORTANT: When you are unsure how to implement something, call these tools FIRST before guessing. They will save you time and help you write correct code.
Available Tools (8 total)
| Tool | What it does | Key parameters |
|---|---|---|
search_knowledge |
Search KB for framework docs, APIs, best practices | query (required), context (optional) |
build_plan |
Build a step-by-step execution plan grounded in KB | goal (required), constraints (optional) |
review_plan |
Review a plan against KB best practices | proposal (required), goal (required) |
verify_code_math |
Verify code against authoritative ML descriptions | code_snippet (required), concept_name (required) |
diagnose_failure |
Diagnose training/deployment failures via KB | symptoms (required), logs (required) |
propose_hypothesis |
Propose ranked next-step hypotheses | current_status (required) |
query_hyperparameter_priors |
Query hyperparameter values and heuristics | query (required) |
get_page |
Retrieve a specific KB page by ID | page_id (required) |
When to Use These Tools
Call search_knowledge for questions like:
- "How do handoffs work in the OpenAI Agents SDK?"
- "How to serialize and restore agent state between requests?"
- "How to disable parallel tool calls when handoff tools are registered?"
- "How to export agent configuration to declarative JSON?"
- "How to build a router agent that classifies into 27 intent categories?"
Call build_plan at the start:
- "Build a multi-agent customer support triage service with 27 intent categories, handoffs, and state persistence using FastAPI"
Call review_plan after drafting your approach:
- Pass your implementation plan and let the KB check it against best practices
Call diagnose_failure when something breaks:
- Pass the error symptoms and logs to get KB-grounded debugging advice
Example Calls
search_knowledge(query="OpenAI Agents SDK handoff mechanism with many specialist agents", context="building multi-agent customer support API with 27 intent categories")
search_knowledge(query="RunState serialization to_json from_json for agent state persistence")
search_knowledge(query="FastAPI SSE streaming with async generators")
build_plan(goal="multi-agent customer support triage API with 27 intent routing, handoffs, state persistence, structured output, and declarative config")
review_plan(proposal="Using OpenAI Agents SDK with handoff() for routing to grouped specialists, fine-grained intent classification in structured output", goal="production customer support triage API with 27 intents")