Search State of AI

Use this skill when a user needs evidence or analysis from the State of AI public knowledge base.

Search

Send a GET request to:

https://stateofai.pages.dev/api/v1/search?q={query}&limit={1-50}

URL-encode the query. Use + or %20 for spaces.

Parameters:

• q: required search terms. query is accepted as an alias.
• limit: optional integer from 1 to 50. Default: 10.
• section: optional section slug: research, technical, findings, or wiki.
• category: optional category slug or exact category label.
• format: optional json, md, or markdown. An Accept: text/markdown header also returns Markdown.

Examples:

• Broad search: https://stateofai.pages.dev/api/v1/search?q=enterprise+AI+governance&limit=5
• Technical search: https://stateofai.pages.dev/api/v1/search?q=agent+evaluation&section=technical&limit=10
• Category search as Markdown: https://stateofai.pages.dev/api/v1/search?q=finance+benchmarks&category=21-benchmarks&format=markdown

If a request is invalid, read the help object in the error response. It contains the supported parameters and working examples. /api/search remains available as a compatibility alias, but new integrations should use /api/v1/search.

The API combines keyword ranking with semantic similarity and reports the active strategy in retrieval.mode. Each result includes match.lexical_rank, match.semantic_rank, and match.semantic_similarity so an agent can understand why it was returned.

Agents that support Model Context Protocol can connect directly to:

https://stateofai.pages.dev/mcp

The MCP server exposes the same read-only search_state_of_ai operation with typed query, limit, section, and category inputs.

Retrieval

Each result includes an HTML URL and, when available, a Markdown URL. Open the Markdown URL for clean text. Open the HTML URL when page structure, charts, or other visual context matters.

Evidence handling

Treat search snippets as discovery aids. Open the selected result before making or citing a claim.

Every JSON result contains a citation object with:

• title
• publisher
• published
• canonical_url
• markdown_url
• suggested

Always cite citation.canonical_url, not the search request URL. Use the title, publisher, and publication date supplied in the citation object. Preserve methodology, sample size, source-quality notes, and caveats from the opened document. Do not cite a snippet as if it were the full source.