Ref takes advantage of MCP sessions to track search trajectory and minimize context usage. There's a lot more ideas cooking but here's what we've implemented so far.

1. Filtering search results

For repeated similar searches in a session, Ref will never return repeated results. Traditionally, you dig farther in to search results by paging to the next result but this approach allows the agent to page AND adjust the prompt at the same time.

2. Fetching the part of the page that matters

When reading a page of documentation, Ref will use the agent's session search history to dropout less relevant sections and return the most relevant 5k tokens. This helps Ref avoid a big problem with standard fetch() web scraping which is when it hits a large documentation page you can easily end up pull in 20k+ tokens into context, most of which are irrelevant.

Why does minimizing tokens from documentation context matter?

1. More context makes models dumber

It's well documented that as of July 2025 that models get dumber as you put in more tokens. You might have heard about how models are great with long context now and that's kind of true but not the whole picture. For a quick primer on some research, checkout this video from the team at Chroma.

2. Tokens cost $$$

Imagine you are using Claude Opus as a background agent and you start by having the agent pull in documentation context and suppose it pulls in 10000 tokens of context with 4000 being relevant and 6000 being extra noise. At API pricing, that 6k tokens cost about $0.09 PER STEP. If one prompt ends up taking 11 steps with Opus, you've spent $1 for no reason.

Setup

There are two options for setting up Ref as an MCP server, either via the streamable-http server (recommended) or local stdio server (legacy).

This repo contains the legacy stdio server.

Streamable HTTP (recommended)

"Ref": {
  "type": "http",
  "url": "https://api.ref.tools/mcp?apiKey=YOUR_API_KEY"
}

stdio

"Ref": {
  "command": "npx",
  "args": ["ref-tools-mcp@latest"],
  "env": {
    "REF_API_KEY": <sign up to get an api key>
  }
}

Tools

Ref MCP server provides all the documentation related tools for your agent needs.

ref_search_documentation

A powerful search tool to check technical documentation. Great for finding facts or code snippets. Can be used to search for public documentation on the web or github as well from private resources like repos and pdfs.

Parameters:

• query (required): Query to search for relevant documentation. This should be a full sentence or question.

ref_read_url

A tool that fetches content from a URL and converts it to markdown for easy reading with Ref. This is powerful when used in conjunction with the ref_search_documentation tool that returns urls of relevant content.

Parameters:

• url (required): The URL of the webpage to read.

OpenAI deep research support

Ref can be used as a source for deep research. OpenAI requires specific tool definitions so when used with an OpenAI client, Ref will provide the same tools with slightly different naming.

ref_search_documentation(query) -> search(query)
ref_read_url(url) -> fetch(id)

Development

npm install
npm run dev

Running with Inspector

For development and debugging purposes, you can use the MCP Inspector tool. The Inspector provides a visual interface for testing and monitoring MCP server interactions.

Visit the Inspector documentation for detailed setup instructions.

To test locally with Inspector:

npm run inspect

Or run both the watcher and inspector:

npm run dev

Local Development

1. Clone the repository
2. Install dependencies:

npm install

3. Build the project:

npm run build

4. For development with auto-rebuilding:

npm run watch

License

MIT