cloro runs a hosted MCP server. Connect Claude, Cursor, or an agent you built, and it can query Google Search, ChatGPT, Perplexity, Gemini, Copilot, and AI Mode as tools, with parsed answers and the sources behind them rather than a training-data guess.
4.7 · 53 reviews
# One command. Your API key is the only credential.
$ claude mcp add --transport http cloro https://mcp.cloro.dev/YOUR_CLORO_API_KEY/mcp
An AI assistant has never seen your SERP. It was trained months ago, on pages rather than result sets, and it cannot check what ChatGPT told a customer about your category this morning. Ask it where you rank and it will answer anyway. The answer is a guess.
Giving it a browser does not fix that. A page fetch returns HTML your agent still has to parse, and it misses most of what decides whether anyone sees you. It cannot tell you which sources an AI engine cited, whether an AI Overview appeared, or what the ads were. cloro's MCP server hands the agent those fields already parsed.
These are real prompts against the tools below. You ask in English and the agent picks the tool.
The answer arrives inside the conversation you are already having. Your agent pulls the data, reasons over it, and writes the summary in one turn, instead of you exporting a CSV and reading it somewhere else.
The flow is short. You connect the client once and ask a question; the agent calls the tool, cloro handles retrieval and parsing, and structured JSON comes back. Each tool maps to one cloro endpoint and costs the same credits it would cost from code.
| Tool | What it returns | Credits |
|---|---|---|
| scrape_google | Google SERP: organic results, ads, and the AI Overview when you ask for it | 3+2/page |
| scrape_chatgpt | ChatGPT's answer, the sources it cited, and its query fan-out | 7 |
| scrape_perplexity | Perplexity's answer with its citation list | 4 |
| scrape_gemini | Gemini's answer and the sources behind it | 4 |
| scrape_copilot | Copilot's answer and its cited pages | 5 |
| scrape_google_ai_mode | Google AI Mode's answer, targetable by location or UULE | 4 |
| scrape_google_news | Google News articles with source, date, and thumbnail | 3+2/page |
| scrape_grok | Grok's answer and sources | 4 |
| list_countries | Valid country codes, filterable by engine | Free |
| list_states | US state codes for sub-country targeting | Free |
Every assistant tool takes a prompt and a country, and accepts a US state for sub-country targeting. scrape_google takes either a query plus a country or a complete google.com/search URL, and can extract the AI Overview and hydrate the AI answers inside People Also Ask. Full schemas are in the MCP documentation.
scrape_grok is registered, but Grok is currently unavailable, so calls to it fail until the provider is restored.
The server speaks Streamable HTTP, so there is nothing to install and no local process or container to run. In Claude Code, one line does it:
claude mcp add --transport http cloro https://mcp.cloro.dev/YOUR_CLORO_API_KEY/mcpIn Claude Desktop or Cursor, paste the JSON block from the top of this page into the app's MCP settings under Settings → Developer. If your client can set custom headers, point it at https://mcp.cloro.dev/mcp and send your key as an Authorization: Bearer header instead. That keeps the key out of a config file you might later share.
The server is a thin wrapper over api.cloro.dev. It holds no business logic and no secrets. It forwards your key on every request, and the API enforces authentication, billing, rate limits, and concurrency. The tool schemas are the schemas the API validates against, so what the agent is told it can send is what the API actually accepts.
MCP is for questions you ask in conversation. Each call is synchronous and counts against your plan's concurrency limit, so it is the wrong tool for volume: a few hundred prompts across six engines will crawl, and you will hit the ceiling before you finish.
For scheduled tracking, such as a prompt set run daily or a keyword list across markets, call the async API directly. It absorbs the concurrency for you. The two fit together: the async API runs the daily job, and MCP answers the questions that job raises.
If you would rather write the integration yourself, the same endpoints are available through the Python and JavaScript SDKs, and through n8n, Zapier, LangChain, and OpenClaw.
Start free. Price per credit drops as your volume grows — see every tier below.
| Plan | Price / mo | Credits | Per 1k | Concurrency | Seats | Support |
|---|---|---|---|---|---|---|
| Free | $0 | 500 | — | 1 | 1 | Documentation assistant |
| Lite | $30 | 37,500 | $0.80 | 10 | Unlimited | |
| Hobby | $100 | 250,000 | $0.40 | 20 | Unlimited | |
| StarterPopular | $250 | 650,000 | $0.39 | 50 | Unlimited | |
| Growth | $500 | 1,350,000 | $0.37 | 75 | Unlimited | Priority email |
| Business | $1,000 | 2,800,000 | $0.36 | 100 | Unlimited | Priority email |
| Enterprise 2K | $2,000 | 5,871,025 | $0.34 | 135 | Unlimited | Slack |
| Enterprise 3K | $3,000 | 9,306,606 | $0.32 | 175 | Unlimited | Slack |
| Enterprise 4K | $4,000 | 12,756,261 | $0.31 | 215 | Unlimited | Slack |
| Enterprise 5K | $5,000 | 16,391,783 | $0.31 | 255 | Unlimited | Slack |
| Enterprise | $5,000+ | Increased concurrency, overages on credits and credit discounts for annual contracts.Know more | ||||
MCP, the Model Context Protocol, is an open standard for handing an AI client a set of tools it can call mid-conversation. An SEO MCP server is one whose tools return search data: rankings, SERP features, or what an AI engine says about a brand. cloro's server exposes its Google SERP API and every AI engine it monitors, so your agent reads live results instead of answering from training data.
Any client that speaks Streamable HTTP. Claude Code connects with a single `claude mcp add` command; Claude Desktop and Cursor take a JSON block in their MCP settings. Clients that can set custom headers can point at `https://mcp.cloro.dev/mcp` and send the key as a Bearer token instead of putting it in the URL path.
No. Each tool call runs one API call at the standard per-provider credit rate and draws on the same credit pool, so connecting over MCP costs exactly what the same request costs from code. See the pricing page for plan rates.
The server holds no secrets and no business logic. It forwards your key to `api.cloro.dev` as a Bearer token, and the API enforces authentication, billing, and rate limits. The URL-path form exists for clients that cannot set headers. If yours can, use the header form and keep the key out of your client's config URL.
No. Every tool call is synchronous and counts against your concurrency limit, so a few hundred prompts across engines will be slow and will hit that ceiling. MCP is for asking questions in conversation. For scheduled tracking across many prompts, call the async API directly.
A general web-search MCP server returns links for an agent to read. cloro returns what a search or AI engine actually showed, parsed into the answer text, the sources it cited, ads, shopping cards, and brand entities. An agent can measure with that, rather than just read. Our Google Search MCP walkthrough covers the setup step by step.