MCP serverOverview
API reference llms.txt
Docs›MCP Server
MCP Server[01]

MCP server

Give any MCP client six Onto tools — read, score, map, batch and extract. Connect by URL with an OAuth sign-in, or run it locally with npx.

What a tool call does

[02]
your client → api.buildonto.dev/mcp
You ask your client about a page

  1. Your agent has no reader of its own for the page — so it looks at the tools it has.

  2. One of Onto's six, with arguments — here read_url and the URL.

  3. Over the connector URL, signed in with OAuth — or through the local npx server with your onto_sk_ key.

  4. Each tool calls its /v1 endpoint, so the credits are the API's, charged once.

  5. The agent reads the clean page and answers from it.

Clean Markdown in the chat, billed like any API call

Install it

[03]
Install in
One command
$ claude mcp add --scope user --transport http onto https://api.buildonto.dev/mcp
Docs →
  1. Run it in a terminal. Claude Code asks you to approve Onto on first use.
  2. Then ask Claude Code to “read example.com with onto”.

Step-by-step for every client is on Set up your client.

Two ways to connect

[04]
Remote connectorLocal server
Add it withThe URL https://api.buildonto.dev/mcpnpx -y @ontosdk/mcp@latest
Signs in withYour Onto account, over OAuthAn onto_sk_ key in ONTO_API_KEY
TransportStreamable HTTP (SSE at /sse too)stdio, Node 20+
Used byClaude Code, Claude Desktop, Cursor, Codex, ChatGPTWindsurf, Cline, Continue — anything that runs a command

Both expose the same six tools and bill the same account. Details on the remote connector and the local server's configuration.

The six tools

[05]
ToolReturnsCallsCredits
read_urlThe page as Markdown, with its size, reduction and cache state/v1/read1
read_and_scoreMarkdown plus the score, grade and hallucination risk/v1/read-and-score1
score_urlJust the score, and why — penalties and recommendations/v1/scorefree
extract_dataThe JSON-LD, OpenGraph and meta the page declares/v1/extract1
map_siteA site's URLs, from its sitemap or its start page's links/v1/map1
batchUp to 50 URLs — or a site — read, scored or extracted/v1/batch1 per URL

Inputs and outputs for each are on the Tools reference.

MCP or the API

[06]

Same endpoints, same credits, same errors — MCP is a transport. Use it when an agent you don't write should read the web; call the Read API yourself when you control the code and want retries, caching and batching on your terms.