Show HN: llm-tools-openapi and MCP: access to MCP server with this one trick

4 days ago 3

a plugins that allow to use OpenAPI as a tools

Install this plugin in the same environment as LLM.

llm install llm-tools-openapi

To use this with the LLM command-line tool:

llm --td --tool 'OpenAPIToolbox(openapi_url="https://your/swagger.json")' 'question your API'

With the LLM Python API:

import llm from llm_tools_openapi import openapi model = llm.get_model("gpt-4.1-mini") result = model.chain( "Example prompt goes here", tools=[openapi] ).text()

To set up this plugin locally, first checkout the code. Then create a new virtual environment:

cd llm-tools-openapi python -m venv venv source venv/bin/activate

Now install the dependencies and test dependencies:

To run the tests:

The llm-tools-openapi plugin is not limited to standard OpenAPI-compatible APIs. You can also use it to interact with any MCP (Model Context Protocol) server by converting it into an OpenAPI server using mcpo.

MCP servers typically communicate over raw stdio, which is not compatible with most tools and lacks standard features like authentication, documentation, and error handling. mcpo solves this by acting as a proxy: it exposes any MCP server as a RESTful OpenAPI HTTP server, instantly making it accessible to tools and agents that expect OpenAPI endpoints.

Benefits:

Instantly compatible with OpenAPI tools, SDKs, and UIs
Adds security, stability, and scalability
Auto-generates interactive documentation for every tool
Uses standard HTTP—no custom protocols or glue code required

1. Start Your MCP Server with mcpo

mcpo also supports serving multiple MCP servers from a single config file. Each tool will be available under its own route, with its own OpenAPI schema and documentation.

Example demo/mcp-demo.json

{ "mcpServers": { "basic-memory": { "command": "uvx", "args": [ "basic-memory", "mcp" ] }, "playwright": { "command": "npx", "args": ["-y", "@playwright/mcp@latest"] } } }

Start mcpo with:

mcpo --config demo/mcp-demo.json

Each tool will be accessible at:

http://localhost:8000/basic-memory
http://localhost:8000/playwright

And their OpenAPI docs at:

http://localhost:8000/basic-memory/docs
http://localhost:8000/playwright/docs

3. Use with llm-tools-openapi

Point the llm-tools-openapi plugin to the OpenAPI endpoint provided by mcpo. For example:

llm \ --td \ --tool 'OpenAPIToolbox(openapi_url="http://0.0.0.0:8000/playwright/openapi.json")' \ --tool 'OpenAPIToolbox(openapi_url="http://0.0.0.0:8000/basic-memory/openapi.json")' \ "Open Hacker News and create a note in News directory For the ten first articles, create a note like this : title - url Number of upvote Number of comment use snapshot and not screenshot" --cl 30

You will obtain a md file with the content you want.

--- title: Hacker News - Top 5 Articles type: note permalink: hacker-news/hacker-news-top-5-articles tags: - '#news' - '#technology' --- 1. **Deep learning gets the glory, deep fact checking gets ignored** [Read more](https://rachel.fast.ai/posts/2025-06-04-enzyme-ml-fails/index.html) Site: fast.ai - Points: 73 - Author: chmaynard - Comments: 5 - Posted: 55 minutes ago 2. **Destination: Jupiter** [Read more](https://clarkesworldmagazine.com/liptak_06_25/) Site: clarkesworldmagazine.com - Points: 52 - Author: AndrewLiptak - Comments: 13 - Posted: 2 hours ago

....

5. **A deep dive into self-improving AI and the Darwin-Gödel Machine** [Read more](https://richardcsuwandi.github.io/blog/2025/dgm/) Site: richardcsuwandi.github.io - Points: 7 - Author: hardmaru - Comments: Discuss - Posted: 1 hour ago

By combining llm-tools-openapi with mcpo, you can: