AI Benchmark Index

Measured conclusion: Alive, not deep-tested. L2 outcome: not_attempted.

Search 264 AI benchmarks by category, with data sources and licences for each.

Add to your agent

Copy this into your MCP client config:

{
  "mcpServers": {
    "AI Benchmark Index": {
      "url": "https://aibenchmarkindex.com/mcp",
      "transport": "streamable-http"
    }
  }
}

Measured tools (6)

• search_listings read-only

  Search published benchmarks by text, place and category.

  Parameters:

  • q — string — Search text, with at least one letter or digit. Matches name, categories, attributes, description and place. Typos fall back to close matches.
  • city — string — City slug, from /places.
  • sort — string — `relevance` needs `q` and is the default with it. `name` is the default without `q`. `recent` is newest first.
  • limit — integer — Items per page.
  • cursor — string — `page.nextCursor` from the previous response. Opaque. Send the same filters and `limit`.
  • region — string — Region slug, from /places.
  • country — string — ISO 3166-1 alpha-2 country code, any case. A country the directory does not cover, as listed in /places, gives 400.
  • category — string — Category slug, from /categories. A parent slug also matches its children.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_listing read-only

  Get one published benchmark by its slug.

  Parameters:

  • slug (required) — string — The listing slug.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_places read-only

  List countries, regions and cities that have benchmarks.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_categories read-only

  List categories with their benchmark counts.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• get_stats read-only

  Get the number of published benchmarks, places and categories, and when the data last changed.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

• list_sources read-only

  List where the benchmark data comes from, with the license and refresh of each source.

  Hints: read-only yes · destructive no · idempotent yes · open-world no

Publisher website Repository

Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.