Data for AI
Captcha MCP connects NeuroCaptcha to Claude, Cursor, VS Code and any other MCP client. When the agent runs into a reCAPTCHA, it calls a tool, gets a token and carries on — on your API key and balance, at the usual API prices.
Data for AI is a family of MCP servers on your NeuroCaptcha account. Captcha MCP works today; the others are in development.
Solves reCAPTCHA v2, v3 and their Enterprise versions for your agent, with or without your proxy. The agent gets a ready token and submits the form itself.
https://mcp.neurocaptcha.comTurns any page into clean Markdown or text that fits the model's context: no markup noise, no scripts.
A cloud browser your agent drives by itself: opens pages, clicks, fills in forms and takes screenshots, with captchas solved along the way.
Extracts structured data — prices, listings, tables — from websites as JSON, following a schema you describe.
Fresh search results with links and snippets, so the agent finds sources before it opens them.
One remote endpoint over Streamable HTTP. Add it to your client together with your API key: no packages, no local process.
MCP endpoint
Streamable HTTP · JSON-RPC 2.0
https://mcp.neurocaptcha.comRun it in a terminal. The server is added to the current project; add --scope user to use it in every project. Type /mcp in Claude Code to check the connection.
claude mcp add --transport http neurocaptcha https://mcp.neurocaptcha.com --header "Authorization: Bearer YOUR_API_KEY"Authorization: Bearer YOUR_API_KEYX-API-Key: YOUR_API_KEY/mcp?key=YOUR_API_KEYinitialize, ping and tools/list answer without a key, so a client can list the tools before you add one. Every tool call needs a key.
Replace YOUR_API_KEY with a key from Dashboard → API keys. A test key returns free fake TEST_ tokens — enough to check the setup.
The agent picks the right tool from its description. Arguments mirror the task fields of the API, so everything in the docs applies.
Solves reCAPTCHA v2 — checkbox, invisible or Enterprise with enterprisePayload — and waits for the token. Optional: your proxy, userAgent, cookies, apiDomain.
{"websiteURL": "https://example.com/login",
"websiteKey": "6Le-SITE-KEY"}{"taskId": 7781, "status": "ready",
"token": "03AGdBq24PBCbwiDRaS_MJ7Z…", "cost": "0.000600"}Gets a reCAPTCHA v3 token for a page action with the score you need. Enterprise is supported; a proxy only for Enterprise.
{"websiteURL": "https://example.com/checkout",
"websiteKey": "6Le-SITE-KEY",
"pageAction": "checkout", "minScore": 0.7}{"taskId": 7782, "status": "ready",
"token": "03AFcWeA6xnQ1bNg…", "cost": "0.000900"}Takes any task object the API accepts and waits for the result. For task types without a dedicated tool.
{"task": {"type": "RecaptchaV2EnterpriseTaskProxyless",
"websiteURL": "https://example.com",
"websiteKey": "6Le-SITE-KEY"}}{"taskId": 7783, "status": "ready",
"solution": {"gRecaptchaResponse": "03AGdBq2…"},
"cost": "0.001500"}Creates a task and returns its taskId right away, without waiting: for parallel work or a callbackUrl.
{"task": {"type": "RecaptchaV2TaskProxyless", …},
"callbackUrl": "https://your.app/hooks/captcha"}{"taskId": 7784, "taskType": "RecaptchaV2TaskProxyless"}Checks a task: processing, or ready with the solution and the cost.
{"taskId": 7784}{"taskId": 7784, "status": "ready",
"solution": {"gRecaptchaResponse": "03AGdBq2…"},
"cost": "0.000600"}Returns the account balance in USD, so the agent can check it before a long run.
{}{"balance": 12.4855, "currency": "USD"}Lists the task types the service accepts right now, with the price per 1,000 and whether a proxy is needed.
{}{"types": [{"type": "RecaptchaV2TaskProxyless",
"pricePer1000": "0.60", "needsProxy": false,
"tool": "solve_recaptcha_v2"}, …]}Reports whether the site accepted or rejected a token. Feeds quality control and refunds, like the report methods of the API.
{"taskId": 7781, "correct": false}{"taskId": 7781, "correct": false, "status": "success"}Captcha MCP has no subscription and no markup: each solved task is charged at the regular price of its type, failed tasks are free.
The server works on your API key: the same balance, rate limits, spending caps and IP allowlist as for direct API calls. Tasks created by the agent appear in your dashboard with the rest.
These are the defaults: lower them per key in the dashboard or ask us to raise them.
Model Context Protocol is an open standard that lets AI applications call external tools. Claude, Cursor, VS Code, Windsurf, ChatGPT and many other clients support it. Captcha MCP is a remote MCP server: you add one URL, nothing runs on your computer.
reCAPTCHA v2 (checkbox and invisible), reCAPTCHA v3 and the Enterprise versions of both, with or without your proxy — the same types as the API. When the API learns new types, the agent gets them through solve_captcha and list_task_types without any change to your setup.
The same as the API: you pay only for solved tasks at the price of their type, and failed tasks are free. The MCP server itself costs nothing and has no subscription.
Create a separate key for agents in the dashboard: limit it with a spending cap and an IP allowlist and revoke it at any time without touching your other integrations. Prefer the Authorization header — the ?key= form puts the key into the URL, so use it only where headers are impossible.
Yes, but check where requests come from. Connectors added in the settings of Claude or ChatGPT connect from the provider's servers, not from your computer. If your key has an IP allowlist, use a separate key for such connectors.
Yes. Test keys work with Captcha MCP too: tasks are free and return a fake token that starts with TEST_ in a couple of seconds. That checks the setup and the agent's flow, but the target site will not accept such a token.
Sign up, create an API key and add Captcha MCP to your client. From the next run, your agent handles captchas by itself.