AI tools leaderboard
Models, agent runtimes, vector stores, evaluation frameworks. The audience is itself agentic; the bar for self-describing capability manifests is highest here.
Not enough scan data yet.
We surface only real, completed public scans. Once enough ai tools domains have a public score on AgentSpeed, this leaderboard fills in automatically.
Scan your site →AI tool sites are scored by AI agents, so the audience is at minimum agent-adjacent. The bar is highest here: an MCP server card at /.well-known/mcp/, a published /llms.txt with the API, the playground, and the SDKs, and explicit support for agent runtimes.
No MCP server card
Agents discover available tools via /.well-known/mcp/server-card.json. Without one, your model does not show up in agent runtimes that support MCP.
Demo playground is unauthenticated UX only
A "try the demo" button that requires session state cannot be cited by agents. Add a documented programmatic equivalent.
Pricing in tokens but no token definition
When pricing is "$X per million tokens" without saying what counts as a token, agents quote misleading per-call costs.
No usage policy / safety page
Agents looking for usage limits or compliance notes (HIPAA, SOC2, GDPR) need a structured page to cite.
- 01Publish /.well-known/mcp/server-card.json describing your MCP tools.
- 02Add Organization + SoftwareApplication JSON-LD on every page.
- 03Document the auth scheme + rate limits + token-counting rules in the SSR HTML.
- 04Publish /llms.txt with Models, API, Playground, Pricing, Safety sections.
- 05Treat agent traffic as primary; test fetches with GPTBot + ClaudeBot UAs.
Each row is the latest completed public scan for one domain. Composite is a 0–100 weighted average of five sub-scores; grade is bound to a published band table. The rubric (current and prior versions) is at /rubric.
How are scores calculated?
Every score is the output of the public AgentSpeed rubric. The full check list, weights, and grade bands are at /rubric. New scans use the current active rubric; historical scans render against their original rubric.
How do I get my domain on the leaderboard?
Visit the scanner and submit a URL to generate a score. Every completed public scan creates a permanent score page and makes the domain eligible for ranking. There’s no application process and no editorial gate.
How do I remove my site?
Add User-agent: AgentSpeedBot / Disallow: / to your robots.txt. On the next scan we honour it automatically and the public score page is replaced with a removal notice. See /bot.
Scan your AI product page to see if agents find your MCP card and policy docs.
Side-by-side composite, sub-scores, and per-category winner. Permanent shareable URL.
Every check in the rubric has a remediation page with paste-ready snippets.
Crawler matrix, llms.txt analyzer, structured-data validator, and seven more.