M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
R

Ref MCP

Community
Token-efficient documentation search for coding agents.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #192 of 438
Stars
★ 1.2k Very popular
Transport
Streamable HTTP · stdio (local process)
Runtime
Node.js · Hosted (no local runtime)
Credentials
API key / credential required
License
MIT
Last commit
Tools
2
55FMRS · C

The supplied materials describe Ref MCP as a token-efficient documentation search and URL-reading server for coding agents, supporting Streamable HTTP and legacy stdio with API-key authentication.

Strongest · Documentation 14/20 Weakest · Security and permissions 8/20

Reliability
9/20
Security and permissions
8/20
Maintenance
11/20
Documentation
14/20
Setup experience
13/20
Why each score
Reliability 9/20
The README identifies two tools and their parameters, while the manifest supplies a Streamable HTTP endpoint, so the normal path is plausible. However, the supplied material contains no committed tests, CI, initialization or tool-list handshake evidence, and does not document error handling, timeouts, dependency failures, or edge cases. The manifest declares no tools, and the README distinguishes remote and legacy stdio implementations, so the score is conservative.
Security and permissions 8/20
The manifest requires an API key in a header marked secret, showing a basic authentication boundary; the documented operations are documentation search and reading, with no visible write or destructive actions. However, the README also shows putting the API key in a URL query parameter, which may expose it through logs or history. Data boundaries, retention, least privilege, and confirmation requirements for private repositories, PDFs, and external URLs are not explained, so points are deducted.
Maintenance 11/20
The repository is not archived, lists an MIT license, has manifest version 3.0.1, and has five open issues, providing a basic maintenance and licensing foundation. The supplied material contains no evidence of commit history, release cadence, dependency updates, issue response, maintainer identity, or a security-response channel, so it cannot receive a sustained-maintenance score near full marks.
Documentation 14/20
The README covers the purpose, two installation modes, authentication examples, both tools and their required parameters, search and read examples, and development and Inspector usage. Points are deducted because the manifest declares no tools, the remote authentication examples are inconsistent between headers and URL parameters, and platform compatibility, quotas, pricing, privacy/data handling, troubleshooting, and service limitations are insufficiently disclosed.
Setup experience 13/20
The README provides Cursor links, Streamable HTTP and stdio configurations, an npx command, and the API-key location, making the basic mainstream-client path reasonably clear. It still requires obtaining an API key, presents inconsistent remote authentication methods, labels stdio as legacy, and provides no broader platform-compatibility or connection-failure evidence, so it does not approach full marks.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • AI coding agents that need technical documentation lookup
  • Teams seeking to reduce irrelevant documentation context and token usage
  • Users who need access to public and authorized private documentation resources

Not for

  • MCP workflows that do not need documentation retrieval
  • Users without a Ref API key
  • General-purpose web scraping scenarios rather than documentation search and reading

Required permissions

  • An API key is required for authentication
  • Private-resource use requires authorization to those private repositories or documents

Risks and side effects

  • The API key is a secret credential and should not be exposed or committed to shared configuration
  • Searching private repositories and PDFs may expose their contents to the agent
  • Retrieved content becomes part of the agent context and should still be checked against the task

Setup

Before you start

Runtime：Node.js · Hosted (no local runtime)

REF_API_KEY requiredsecret API key for the Ref documentation search service; sign up at ref.tools to obtain one.

Streamable HTTP is recommended: connect to https://api.ref.tools/mcp and provide an API key. The legacy stdio option runs npx ref-tools-mcp@latest with the API key supplied through REF_API_KEY.

.cursor/mcp.json 
{"Ref":{"command":"npx","args":["ref-tools-mcp@latest"],"env":{"REF_API_KEY":"YOUR_API_KEY"}}}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "Ref": {
    "command": "npx",
    "args": [
      "ref-tools-mcp@latest"
    ],
    "env": {
      "REF_API_KEY": "YOUR_API_KEY"
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Generate and validate client config

Check that it works

Check that ref_search_documentation and ref_read_url appear in your client's tool list, then ask the agent to search a simple API question; a search result with token counts and doc URLs confirms the connection.

Troubleshooting

  1. Verify that the API key is valid and supplied through X-Ref-Api-Key, the URL configuration, or REF_API_KEY
  2. For Streamable HTTP, check the https://api.ref.tools/mcp endpoint and authentication settings
  3. For stdio, confirm that npx is available and ref-tools-mcp@latest can be run
  4. If retrieved content is unexpected, search first with ref_search_documentation and then call ref_read_url on a relevant URL

Things to try

Once connected, you can ask your AI assistant things like:

  • Find documentation for Figma's Comment REST API for posting a comment and link the page.
  • Search for best practices on using n8n's merge node versus the Code node with multiple inputs.
  • Read https://docs.n8n.io/code/code-node/ and summarize the Usage section.

Tools 2

ref_search_documentation read-only
Searches technical documentation for facts or code snippets across public web and GitHub content, as well as private repositories and PDFs.
ref_read_url read-only
Fetches a URL and converts its content to Markdown, useful for reading documentation pages returned by search.

Use cases

Find API, service, and library documentation for coding agents
Search for relevant documentation and then read selected pages
Retrieve facts and code snippets from public or private resources

Supported clients

Cursor

Listed from the project's documentation, not tested by this site.

Overview

Ref MCP is a Model Context Protocol server that gives AI coding tools and agents access to public and private documentation for APIs, services, and libraries. It filters repeated search results and returns the most relevant parts of pages using session search history to reduce context usage.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision f5feb829cc7b Data synced 2026-10-19 Read the FMRS scoring method