# 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. ```bash 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 416+ 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: ```bash 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. 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 ```bash claude mcp add clirank -- npx -y clirank-mcp-server@latest ``` Or add to `~/.claude.json` manually: ```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: ```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: ```json { "servers": { "clirank": { "type": "stdio", "command": "npx", "args": ["-y", "clirank-mcp-server@latest"] } } } ``` ### Cline / Continue / Windsurf Use the same stdio command: ```bash 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: - https://clirank.dev/tasks/llm-api-for-coding-agents - https://clirank.dev/tasks/authentication-api-for-ai-agents - https://clirank.dev/tasks/database-api-for-ai-agents - https://clirank.dev/tasks/payments-api-for-ai-agents - https://clirank.dev/tasks/transactional-email-api-for-agents - https://clirank.dev/tasks/deployment-api-for-ai-agents - https://clirank.dev/tasks/image-generation-api-for-agents ## 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 ```bash 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.