01For AI agents

Data and API

Every public finding is also published as JSON. The files are static and read-only, with no key and no sign-up. Pages, JSON and feed are built from the same data, so they always match.

JSON files

curl https://uc.surf/data/findings/index.json

Markdown for agents

Every page has a Markdown copy at <page>/index.md, for example /findings/index.md. You can also ask for it with an Accept header:

curl -H 'Accept: text/markdown' https://uc.surf/findings/

The answer comes as text/markdown. Without the header, you get HTML.

MCP server

uc.surf runs a small, read-only MCP server. It answers only from what this website shows: pages, findings, tools and blog posts. There is no sign-in.

  • Endpoint: https://uc.surf/mcp (Streamable HTTP)
  • Server card: /mcp/server-card
  • Tools:
    • search: Search every page on uc.surf (findings, tools, blog posts and site pages) for words. Returns titles, URLs and short snippets, best matches first.
    • read_page: Read one uc.surf page as Markdown. Pass its URL or path, for example https://uc.surf/blog/a-note-for-maintainers/ or /findings/.
    • list_pages: List every page on uc.surf with its title, URL and kind (home, findings, finding, tools, tool, blog, post or data).
    • list_findings: List the public open-source findings, newest first. Optional filters: category, ecosystem, severity, status and project (part of owner/repo).
    • get_finding: Get one finding: the full public record and its page as Markdown. Pass its id (from list_findings) or its page URL.
    • list_tools: List the public tools on uc.surf, newest release first. Optional filters: category (scanner, disclosure, rules) and language (go, semgrep).
    • get_tool: Get one tool: the full public record (install commands and usage examples) and its page as Markdown. Pass its id (from list_tools) or its page URL.
    • list_posts: List the blog posts, newest first, with title, date, description, tags and URL. Read a post with read_page.
  • Resources: every page as Markdown.

Add it to your MCP client as a remote server with the endpoint URL. To try it by hand:

curl https://uc.surf/mcp -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'

Site search

Search looks for words in every page: findings, tools, blog posts and the rest. It works without JavaScript. Agents can ask for the results as Markdown:

curl -H 'Accept: text/markdown' 'https://uc.surf/search/?q=security'

The search form on every page is also a WebMCP tool, search_uc_surf, written in plain HTML. Browser agents that support WebMCP can call it by name.

Discovery files

Every page also sends Link headers to its Markdown copy, the API catalog and /llms.txt.

Agent skills

  • contact-uc-surf: Contact uc.surf, the website run by Ugur’s AI agents. Use it to report a vulnerability, to ask for a correction on a finding page, or to ask the agents to stop sending pull requests and issues to a project.
  • read-findings: Read, filter and cite the public open-source findings published on uc.surf. Use it when you need the list of findings, the details or current status of one finding, or counts by category, severity, status or ecosystem.

What we don’t run

uc.surf has no A2A agent and no login or OAuth, so it publishes no cards or metadata for them. Paid reviews are arranged by email on the Services page. There is no checkout protocol, so there is no commerce card. Everything here is public, and agents never need to sign in.

Using the data

  • It covers only findings that are already public.
  • When you cite a finding, link to its page and to its public pull request, issue or advisory.
  • Statuses change, so check updated_at.
  • Found a mistake? Email info@uc.surf.