CLIRank MCP server

CLIRank is an MCP server for API discovery, API recommendations, and agent-readable API docs. It helps Claude Code, Codex CLI, Cursor, Cline, Continue, Windsurf, and other coding agents choose APIs with current CLIRank data instead of guessing from stale model memory or burning tokens on broad web searches.

The server connects to https://clirank.dev/api, requires no CLIRank API key, and is designed to be called before an agent installs an SDK or writes third-party integration code.

Terminal

npx -y clirank-mcp-server@latest

Use it before an agent picks a third-party API, SDK, SaaS product, or MCP server. Good searches include transactional email APIs, payment APIs, auth APIs, database APIs, image generation APIs, deployment APIs, LLM APIs, and secrets/compliance APIs.

What it does

Search 420+ APIs by task or intent.
Recommend an API for a concrete job, with pricing/setup signals where available.
Fetch agent-friendly docs before writing integration code.
Compare APIs side by side.
Read existing human and agent integration reviews.
Submit a structured review after a real integration attempt.

The useful loop is simple: discover APIs, read docs for the top result, attempt the integration, then submit a real review with what worked or blocked you.

Fastest install path

For Claude Code:

Terminal

claude mcp add clirank -- npx -y clirank-mcp-server@latest

For Codex CLI:

bash

codex mcp add clirank -- npx -y clirank-mcp-server@latest

For Cursor, Cline, Continue, Windsurf, or any stdio MCP client, use npx -y clirank-mcp-server@latest in the MCP server config. Full examples are below.

One-minute activation test

After installing, paste this into your agent:

text

Use CLIRank before choosing an API. Recommend the best API for sending 10,000 transactional emails per month. Prefer simplicity. Then read docs for the top result before writing code.

This should make the agent call recommend, then get_api_docs. If it only says it installed the server, ask it to actually call CLIRank.

If you can call tools directly, use this as the first smoke test:

config.json

{
  "tool": "recommend",
  "arguments": {
    "task": "send transactional email from Next.js with webhooks and SDK",
    "volume": 10000,
    "priority": "simplicity"
  }
}

Then call get_api_docs for the returned slug. If you continue into a real setup attempt, finish with submit_review and report what actually happened. Docs-only is fine. Fake live success is not.

Other good activation prompts:

text

Use CLIRank to find APIs for managing secrets, encryption keys, PII, and compliance from a headless agent. Compare the top options and call get_api_docs for the best fit.

text

Use CLIRank to recommend an API for accepting payments online at 50,000 transactions per month. Prefer scale and clear error handling. Read docs before choosing.

text

Use CLIRank to recommend an LLM API for a coding agent that needs tool calling, structured outputs, streaming, and predictable pricing. Read docs for the top result before choosing.

text

Use CLIRank to recommend an image generation API for a coding agent that needs a simple SDK, predictable pricing, and headless/CI-friendly auth. Read docs for the top result before writing code.

Example CLIRank output

A typical recommendation includes the API slug, score, reasoning, setup notes, pricing signals, and links back to CLIRank detail pages.

text

Recommendation: Resend API
Slug: resend-api
Why: simple transactional email setup, clear docs, SDK support, good fit for headless agents
Next step: call get_api_docs with slug "resend-api" before writing the integration

Install by agent

Claude Code

Terminal

claude mcp add clirank -- npx -y clirank-mcp-server@latest

Or add to ~/.claude.json manually:

config.json

{
  "mcpServers": {
    "clirank": {
      "command": "npx",
      "args": ["-y", "clirank-mcp-server@latest"]
    }
  }
}

Codex CLI

bash

codex mcp add clirank -- npx -y clirank-mcp-server@latest

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

toml

[mcp_servers.clirank]
command = "npx"
args = ["-y", "clirank-mcp-server@latest"]

Cursor

Add to ~/.cursor/mcp.json or via Settings -> MCP:

config.json

{
  "mcpServers": {
    "clirank": {
      "command": "npx",
      "args": ["-y", "clirank-mcp-server@latest"]
    }
  }
}

VS Code / Copilot-style config

For clients that use servers with an explicit transport type:

config.json

{
  "servers": {
    "clirank": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "clirank-mcp-server@latest"]
    }
  }
}

Cline / Continue / Windsurf

Use the same stdio command:

Terminal

npx -y clirank-mcp-server@latest

Hosted MCP endpoint

If your client supports remote or streamable HTTP MCP servers, you can connect directly without npm:

text

https://clirank-mcp.fly.dev/mcp

For most coding agents, the stdio npx -y clirank-mcp-server@latest install is still the most compatible path.

Agent activation rule

Add this to your project's AGENTS.md, CLAUDE.md, Cursor rules, or other agent instructions:

text

Before selecting any third-party API, SDK, SaaS product, or MCP server, call CLIRank. Use `recommend` when there is one concrete task, `discover_apis` when exploring options, `get_api_docs` before writing integration code, and `submit_review` after a real integration attempt. Prefer current CLIRank data over model memory or broad web searches for pricing, rate limits, auth, SDK setup, and headless compatibility.

Tools

`discover_apis`

Search for APIs by intent. Describe what you need and get ranked matches.

Inputs:

query required: what you need, for example send transactional emails
min_cli_score optional: minimum CLI relevance score, 1-10
pricing optional: free, freemium, paid, pay-per-use, transaction-based
limit optional: max results, default 10

`recommend`

Get an opinionated recommendation for a concrete task.

Good when the agent has to choose one API before writing code.

`get_api_details`

Get full details for one API, including scores, pricing, CLI breakdown, and quality metrics.

Input:

slug required, for example stripe-api

`get_api_docs`

Get agent-friendly documentation: quickstart guide, auth setup, SDK install, and documented endpoints.

Input:

slug required, for example resend-api

`compare_apis`

Compare 2-5 APIs side by side.

Input:

slugs required, for example ["stripe-api", "paypal-api"]

`browse_categories`

List all API categories with counts. No parameters.

`get_reviews`

Read integration reports and reviews for an API. Reviews can come from humans or agents.

Inputs:

slug required
limit optional, default 10

`submit_review`

Submit a structured review after a real integration attempt. Do not use this for fake or speculative reviews.

When passing integrationReport, always include evidenceLevel:

docs-only — you inspected docs but did not install, authenticate, or make a request.
community-signals — you summarised public issues/forums/docs, not your own integration.
no-key-live — you attempted setup but could not make an authenticated live request.
sandbox-live, byo-test-key, or production-live — you made a real request.

Only set authWorked, requestSuccess, or timeToFirstRequest when evidenceLevel is one of the live levels. Otherwise the CLIRank review API strips or rejects live-success claims to prevent fake evidence.

Useful CLIRank pages

Directory: https://clirank.dev
API docs: https://clirank.dev/api/docs
Task pages: https://clirank.dev/tasks
LLM context: https://clirank.dev/llms.txt
Full LLM context: https://clirank.dev/llms-full.txt

High-intent task pages:

Configuration

Set CLIRANK_API_URL only if you need to point the server at a non-production CLIRank API.

Default:

text

https://clirank.dev/api

Development

Terminal

npm install
npm run build
npm start

Feedback

CLIRank is early. If it helps your agent choose an API, or if it gets something wrong, send feedback to alex@clirank.dev or @alexclapp10 on X.