For AI agents

Let your AI agent make your website easy to find and ready for AI answers.

Use Claude, ChatGPT or another AI assistant? Connect it to Onbiom once. Then ask it in your own words to check your website and prepare fixes. Nothing on your website changes until you approve it and put it live.

Connect your assistant

You need a free Onbiom account. In Claude.ai or Claude Desktop:

  1. Open Settings, then Connectors, and add a custom connector.
  2. Paste this address: https://onbiom.com/mcp
  3. Sign in to Onbiom and choose what your assistant may do.

Then ask: “Check my website and tell me what to fix first.” Other assistants are covered below.

From check to verified change: what your assistant does for you

Four steps. Your assistant does the legwork; only you approve.

  1. Check. Your assistant checks your public website. It tells you the business details it found, where each came from, and what is missing or conflicting.
  2. Draft. Once you allow it, the assistant writes the missing details into a private draft that expires after 24 hours. It asks you one question about anything it could not find, and keeps every edit as a new version you can read back.
  3. Approve. The assistant gives you a link that works once and expires in 15 minutes. Open it signed in to Onbiom, read the exact version and approve it yourself. An approval covers that one version; any later edit needs a new approval. Approving does not change your website.
  4. Confirm. After you or your developer put the approved changes on your website, the assistant asks Onbiom for a fresh check. You see which details matched, which differed and which we could not see. A match means we saw the detail on your public page once; it does not prove Google or an AI assistant has picked it up.

Nothing in these steps publishes, installs or changes anything on your website. Drafting, approving and confirming cost no Onbiom credits.

Example questions to ask

Other assistants

We have not yet tested these setup screens in each assistant ourselves, so menu names may differ.

ChatGPT

Depends on your plan and workspace settings. Open Settings, then Apps (or Plugins), turn on developer mode if asked, add a custom connector (ChatGPT may call it an MCP server) with https://onbiom.com/mcp, sign in and let it load the tools. Onbiom is not listed in any ChatGPT directory.

Claude Code

Run claude mcp add --transport http onbiom https://onbiom.com/mcp, then sign in from the /mcp menu.

Codex

Add this to ~/.codex/config.toml, then sign in the way your Codex version offers (often codex mcp login onbiom):

[mcp_servers.onbiom]
url = "https://onbiom.com/mcp"

Assistants that only accept a fixed key can use a personal key from Account, Connect your AI client. Personal keys cover monitors only, expire after 30, 60 or 90 days (you choose, default 90) and can be turned off or renamed there.

What still needs your own click in Onbiom

Your assistant can never approve these for you. Each needs a separate step in the Onbiom web app:

For developers: tools, sign-in and data

Onbiom connects to assistants through MCP, the open standard AI assistants use to talk to outside services. Your assistant picks the tools below itself; you never type their names.

Three ready-made prompts guide the steps above: check a website and prepare changes, review one version, and check that approved changes are live. Two read-only resource templates let your assistant open one draft version or one saved check result you own. Prompts and resources approve nothing.

For people who run their own MCP server, read tools return your monitor names, endpoint addresses, sampled connection results, availability figures and, with separate consent, saved website findings and plan allowances. One write tool, add_monitor, adds a public HTTPS endpoint without credentials and schedules inventory checks. Pause and resume need a separate scheduling permission. There are no tools to delete monitors, run tools, buy plans or publish pages. Results are sampled evidence, not a guarantee that an endpoint or tool is correct or always available.

Every tool needs you to sign in with Google or email and approve the permissions it asks for. Signing in alone gives your assistant nothing. Sign-in uses the standard secure sign-in for AI assistants: you sign in on Onbiom, not inside the assistant, so your password never reaches it. You see the exact permissions and approve them, and you can disconnect the assistant at any time in Account, AI client connections.

Data handling: your assistant gets only what the tools you approve need. Raw tool inputs and results, email addresses, billing information and credentials are left out of tool responses. Onbiom does not log tool inputs or results from assistant calls beyond the evidence it saves for your monitors; saved evidence is kept for your plan's retention period, and other records follow the privacy page. Monitor and tool names can still be sensitive, so connect only assistants you trust. Access lasts one hour per token and at most 30 days per connection; you can disconnect at any time.

Support: support@onbiom.com.

How MCP connection monitoring works →

Start free Current plans