leeroopedia-mcp 0.1.6

MCP server for Leeroopedia ML/AI knowledge search

Leeroopedia MCP Server

Give your AI coding agent access to best-practices of ML and AI.

---

$20 free credit on sign-up : that's plenty of searches, plans, and diagnoses. Skip the guesswork on your next fine-tuning run or inference deployment. No credit card required. Get your API key →

What is Leeroopedia?

Your ML & AI Knowledge Wiki. Learnt by AI, built by AI, for AI.

Expert-level knowledge across the full ML & AI stack, from fine-tuning and distributed training, to inference serving and GPU kernel optimization, to building agents and RAG pipelines. 1000+ frameworks and libraries, all in one place.

This MCP server turns your AI coding agent (Claude Code, Cursor, OpenAI Codex, ...) into an ML/AI expert engineer.

Browse the full knowledge base at leeroopedia.com.

Want to go end-to-end?

Leeroopedia gives your agent the knowledge. Kapso gives it the ability to act on it : research, experiment, and deploy. Together: a complete ML/AI engineer agent.

Benchmarks

We measured the effect of Leeroopedia MCP on real ML tasks built by Claude Code.

• ML Inference Optimization: Write CUDA/Triton kernels for 10 KernelBench problems. 2.11x geomean speedup vs 1.80x (+17%), with/without Leeroopedia MCP. → results

• LLM Post-Training: End-to-end SFT + DPO + LoRA merge + vLLM serving + IFEval on 8×A100. 21.3 vs 18.5 IFEval strict-prompt accuracy, 34.6 vs 30.9 strict-instruction accuracy, 272.7 vs 231.6 throughput. → results

  

• Self-Evolving RAG: Build a RAG service that automatically improves itself over multiple rounds. 45.16 vs 40.51 Precision@5, 40.32 vs 35.29 Recall@5, in 52 vs 62 min wall time. → results

  

• Customer Support Agent: Multi-agent triage system classifying 200 tickets into 27 intents. 98 vs 83 benchmark performance, 11s vs 61s per query. → results

  

Quick Start

1. Get Your API Key

1. Go to app.leeroopedia.com
2. Create an account or log in
3. Navigate to Dashboard > API Keys
4. Copy your API key (format: kpsk_...)

2. Use the hosted server (no installation needed)

Just paste this URL into any MCP client that supports remote servers:

https://mcp.leeroopedia.com/mcp?token=kpsk_your_key_here

See the connect guides for all IDEs, running locally, and troubleshooting.

3. Configure Claude Code

Remote (no install):

claude mcp add --transport http leeroopedia "https://mcp.leeroopedia.com/mcp?token=kpsk_your_key_here"

Local (via uvx):

Add to your ~/.claude.json or project .mcp.json:

{
  "mcpServers": {
    "leeroopedia": {
      "command": "uvx",
      "args": ["leeroopedia-mcp"],
      "env": {
        "LEEROOPEDIA_API_KEY": "kpsk_your_key_here"
      }
    }
  }
}

Getting spawn uvx ENOENT? Your IDE can't find uvx in its PATH. Run which uvx (or where uvx on Windows) in your terminal to get the full path, then use it in your config:

"command": "/home/username/.local/bin/uvx"

Common locations: ~/.local/bin/uvx (Linux), ~/.local/bin/uvx (macOS curl install), /opt/homebrew/bin/uvx (macOS Homebrew).

4. Configure Cursor

Remote (no install):

Add to your Cursor settings (.cursor/mcp.json):

{
  "mcpServers": {
    "leeroopedia": {
      "url": "https://mcp.leeroopedia.com/mcp?token=kpsk_your_key_here"
    }
  }
}

Local (via uvx):

{
  "mcpServers": {
    "leeroopedia": {
      "command": "uvx",
      "args": ["leeroopedia-mcp"],
      "env": {
        "LEEROOPEDIA_API_KEY": "kpsk_your_key_here"
      }
    }
  }
}

Getting spawn uvx ENOENT? See the tip in Configure Claude Code above.

5. Configure OpenAI Codex

Run the CLI command:

codex mcp add leeroopedia --env LEEROOPEDIA_API_KEY=kpsk_your_key_here -- uvx leeroopedia-mcp

Or add to your ~/.codex/config.toml:

[mcp_servers.leeroopedia]
command = "uvx"
args = ["leeroopedia-mcp"]

[mcp_servers.leeroopedia.env]
LEEROOPEDIA_API_KEY = "kpsk_your_key_here"

Getting spawn uvx ENOENT? See the tip in Configure Claude Code above.

Available Tools

The MCP server provides 8 agentic tools. Each tool (except get_page) triggers an AI agent on the backend that searches the knowledge base from multiple angles, reads relevant pages, and synthesizes a structured response.

Search & Retrieve

search_knowledge: Search the KB for framework docs, APIs, and best practices

An AI agent synthesizes a grounded answer with [PageID] citations.

Parameter	Required	Description
query	Yes	What you want to find out
context	No	Optional context about what you're building

get_page: Retrieve a specific KB page by ID

Direct lookup, no AI agent needed. Use this to drill into [PageID] citations from other tools.

Parameter	Required	Description
page_id	Yes	Exact page ID (e.g., Workflow/QLoRA_Finetuning, Principle/LoRA_Rank_Selection)

Plan & Review

build_plan: Build a step-by-step ML execution plan

Returns an overview, key specs, numbered steps, and validation criteria, all grounded in KB evidence.

Parameter	Required	Description
goal	Yes	What you want to accomplish
constraints	No	Constraints or requirements (e.g., hardware limits, time budget)

review_plan: Review a plan against KB best practices

Catches incorrect assumptions before you write code. Returns approvals, risks, and improvement suggestions.

Parameter	Required	Description
proposal	Yes	The plan or proposal to review
goal	Yes	The intended goal of the plan

Verify & Debug

verify_code_math: Verify code against ML/math concepts

Checks your code against documented behavior and reference implementations. Returns a Pass/Fail verdict with analysis.

Parameter	Required	Description
code_snippet	Yes	The code to verify
concept_name	Yes	The mathematical/ML concept being implemented

diagnose_failure: Diagnose training/deployment failures

Matches symptoms against known failure patterns and misconfigurations. Returns diagnosis, fix steps, and prevention advice.

Parameter	Required	Description
symptoms	Yes	Description of the failure symptoms
logs	Yes	Relevant log output or error messages

Explore & Optimize

propose_hypothesis: Propose ranked next-step hypotheses

When you're stuck, get alternative approaches ranked by fit, backed by documented patterns. Returns ranked ideas with rationale and suggested experiments.

Parameter	Required	Description
current_status	Yes	Where the project stands now
recent_experiments	No	Description of recent experiments and their outcomes

query_hyperparameter_priors: Query hyperparameter values, ranges & heuristics

Start with battle-tested defaults instead of guessing. Returns a suggestion table with KB-grounded justification.

Parameter	Required	Description
query	Yes	Hyperparameter question (e.g., "learning rate for LoRA fine-tuning Llama-3 8B")

Documentation

docs.leeroopedia.com — full setup guides, environment variables, troubleshooting, connect guides for every IDE, and detailed tool usage.

Contributing

We welcome contributions! Please see our Contributing Guide for details on how to get started.

This project follows our Code of Conduct.

License

This project is licensed under the MIT License.