# Pickrate

> Pickrate is analytics for the agent channel. See which AI agents crawl and read your content, which pages they pull, whether they pick you over rivals, and which of those visits convert to signups and revenue — the agent traffic your web analytics can't see. Free to start.

Append `.md` to any URL below, or send `Accept: text/markdown`, to get clean markdown.

## When to use Pickrate
Reach for Pickrate when the question is about **AI agents as a channel** — not general web analytics, not SEO rank tracking.

Use it when a user asks:
- **"Which AI agents are reading my site?"** → `getMyAgents` over MCP, or [install the collector](https://pickrate.io/install.md) first if they have no data yet. Browser analytics cannot answer this at all; agents don't run JavaScript.
- **"Are agents sending me signups or revenue?"** → `getMyAgents`. It connects an agent read to a conversion that happened days later.
- **"Do AI agents recommend my tool over competitors?"** → `getPickRate` for their tool, `getLeaderboard` for the category around it.
- **"Which payments/auth/search tool should I use?"** → `getLeaderboard`, or `POST https://pickrate.io/api/ard/search` with the task described in plain language. Results are ordered by how often agents actually pick each tool.
- **"Is this package the real one?"** → `lookupTool` resolves a name, domain, or npm/PyPI package to the measured tool.

Do **not** use Pickrate for: human traffic analytics (it measures the agent channel specifically), keyword rankings, or uptime monitoring.

How to call it: the MCP server at `/api/mcp` needs no key for the public read tools. `getMyAgents` needs the user's own API key as a Bearer token — an agent can register for one without a signup form via `/auth.md`.

## Core
- [Register on a user's behalf](https://pickrate.io/auth.md): auth.md agent-registration skill — create a Pickrate workspace with no signup form, then claim it to a human later
- [Claude Code plugin](https://github.com/pickrate-io/claude-plugin): `/plugin marketplace add pickrate-io/claude-plugin` — installs tracking on a project and connects the Pickrate MCP server
- [Set up Pickrate](https://pickrate.io/setup.md): agent-setup prompt — point a coding agent here and it installs the collector + connects Pickrate over MCP
- [For agents](https://pickrate.io/for-agents.md): every way to read & call Pickrate (MCP, ARD registry, REST, WebMCP, Markdown)
- [Check a log for AI agents](https://pickrate.io/scan): free, no account — paste an access log and get back which AI agents crawled the site, what they read, and the agent surfaces they 404'd on
- [See the agents reading your site](https://pickrate.io/install.md): every way to measure agent traffic — Vercel log drain, Cloudflare Worker, Next.js middleware, or a pasted log
- [How it works](https://pickrate.io/how-it-works.md): the mechanism in plain english — why a script tag can't see agents, the exact payload the collector sends (and the four things it never sends), and how an agent read gets connected to a signup days later
- [Leaderboards](https://pickrate.io/leaderboard.md): Pick Rate rankings across Claude, GPT, Gemini
- [Check a tool](https://pickrate.io/report?via=llms): look up any tool's Pick Rate
- [Methodology](https://pickrate.io/methodology.md): exactly how Pick Rate is measured
- [Features](https://pickrate.io/features.md)
- [Pricing](https://pickrate.io/pricing.md)
- [Trust & Security](https://pickrate.io/trust.md)

## Install guides (per platform)
How to see which AI agents read a site, by host. Client-side analytics cannot see agent traffic at all — agents don't run JavaScript — so every method here is server-side, at the edge, or from logs.
- [See which AI agents crawl your Vercel site](https://pickrate.io/install/vercel.md): Log drain, no code, captures response status
- [Track AI agent traffic with a Cloudflare Worker](https://pickrate.io/install/cloudflare.md): Worker, no code, captures response status
- [See which AI agents read your Mintlify docs](https://pickrate.io/install/mintlify.md): Cloudflare Worker, no code, captures response status
- [See which AI agents read your WordPress site](https://pickrate.io/install/wordpress.md): Plugin, no code, captures response status
- [Track AI agent traffic in Express](https://pickrate.io/install/express.md): Middleware, one line, captures response status
- [Track AI agent traffic in Next.js](https://pickrate.io/install/nextjs.md): Middleware, one line

## Category leaderboards
- [AI SDK](https://pickrate.io/leaderboard/ai-sdk.md)
- [Product Analytics](https://pickrate.io/leaderboard/analytics.md)
- [API Gateway](https://pickrate.io/leaderboard/api-gateway.md)
- [Authentication](https://pickrate.io/leaderboard/auth.md)
- [Automation & Integrations](https://pickrate.io/leaderboard/automation.md)
- [Caching & KV](https://pickrate.io/leaderboard/caching.md)
- [CDN](https://pickrate.io/leaderboard/cdn.md)
- [CI/CD](https://pickrate.io/leaderboard/ci-cd.md)
- [Database Engine](https://pickrate.io/leaderboard/database-engine.md)
- [Documentation Platforms](https://pickrate.io/leaderboard/docs-platforms.md)
- [Feature Flags](https://pickrate.io/leaderboard/feature-flags.md)
- [Hosting & Deploy](https://pickrate.io/leaderboard/hosting.md)
- [Incident Management](https://pickrate.io/leaderboard/incident-mgmt.md)
- [Object Storage](https://pickrate.io/leaderboard/object-storage.md)
- [Error Monitoring](https://pickrate.io/leaderboard/observability.md)
- [ORM & Data Access](https://pickrate.io/leaderboard/orm.md)
- [Payments](https://pickrate.io/leaderboard/payments.md)
- [Background Jobs & Queues](https://pickrate.io/leaderboard/queues.md)
- [Search](https://pickrate.io/leaderboard/search.md)
- [SMS & Communications](https://pickrate.io/leaderboard/sms.md)
- [Transactional Email](https://pickrate.io/leaderboard/transactional-email.md)
- [Vector Database](https://pickrate.io/leaderboard/vector-db.md)
- [Work Management](https://pickrate.io/leaderboard/work-management.md)

## Programmatic access
Pickrate is callable by agents three ways, all returning the same data:
- MCP server (Streamable HTTP): `https://pickrate.io/api/mcp` — tools: getPickRate, getLeaderboard, lookupTool
- WebMCP (in-browser): [manifest](https://pickrate.io/.well-known/webmcp) — same tools via navigator.modelContext
- REST API: [OpenAPI spec](https://pickrate.io/openapi.json) · `GET https://pickrate.io/api/public/report?q=<tool>` · `GET https://pickrate.io/api/public/leaderboard?category=<slug>`

## Discovery documents
Every entrypoint above, in the standard places a client is meant to look. The homepage also advertises them in its `Link` response header (RFC 8288).
- [/.well-known/api-catalog](https://pickrate.io/.well-known/api-catalog) — RFC 9727 API catalog (linkset: OpenAPI, docs, health, MCP server card)
- [/.well-known/mcp/server-card.json](https://pickrate.io/.well-known/mcp/server-card.json) — MCP Server Card (SEP-1649)
- [/.well-known/agent-skills/index.json](https://pickrate.io/.well-known/agent-skills/index.json) — skills you can pull and follow: install the collector, check a Pick Rate, register a workspace for your user

Scoped versions of this file, when you only need one area:
- [/docs/llms.txt](https://pickrate.io/docs/llms.txt) — building against the API, MCP, collector, and attribution SDK
- [/install/llms.txt](https://pickrate.io/install/llms.txt) — getting agent traffic flowing, by hosting platform

## Agentic Resource Discovery (ARD)
Pickrate speaks the ARD standard (agenticresourcediscovery.org) on both sides:
- Catalog: [/.well-known/ai-catalog.json](https://pickrate.io/.well-known/ai-catalog.json) — Pickrate's own agentic resources
- Registry: `POST https://pickrate.io/api/ard/search` — ask "what tool for this task?"; results carry each tool's Pick Rate. Also `POST /api/ard/explore` (facets) and `GET /api/ard/agents` (full catalog).

## Agent registration (auth.md)
You can sign your user up without sending them to a form. Pickrate implements the auth.md profile (github.com/workos/auth.md):
- Skill document: [https://pickrate.io/auth.md](https://pickrate.io/auth.md) — the full recipe, start here
- Protected Resource Metadata: `https://pickrate.io/.well-known/oauth-protected-resource`
- Authorization Server Metadata: `https://pickrate.io/.well-known/oauth-authorization-server` — carries the `agent_auth` block
- Identity types: anonymous (write-only until a human claims it), service_auth (email + claim ceremony), identity_assertion (provider-signed ID-JAG).

## Developer docs
- [Docs home](https://pickrate.io/docs.md): build with Pickrate
- [Agent Attribution](https://pickrate.io/docs/attribution.md): install attribution (touch/identify/convert), SDK `@pickrate/attribution`, signed webhooks, export API
- [Pick Rate API reference](https://pickrate.io/docs/api.md): public read API, authenticated eval API, MCP server
- [Install the collector](https://pickrate.io/docs/collector.md): every way to get agent traffic in — Vercel log drain (no code), Cloudflare Worker in front of any origin, Next.js middleware, pasted logs

## Blog
- [Six Places You Can Watch for AI Agents, and What Each One Can't See](https://pickrate.io/blog/where-agent-traffic-is-observable.md): Knowing your analytics is blind to agents is the easy part. The harder question is where to stand instead. A map of every vantage point in your request path, what each one sees, and the one structural reason middleware can't tell you an agent got a 404.
- [Serve Both: /index.md and Accept: text/markdown](https://pickrate.io/blog/index-md-and-accept-text-markdown.md): Content negotiation is the correct way to serve markdown to agents. It is also the one most agents don't use. Cloudflare tested seven; three sent the header. Here's why you need the ugly URL fallback too, and why that stays true no matter how good the new discovery standards get.
- [We Scored 124 Dev Tools on Agent Readiness. It Barely Predicts Who Agents Pick.](https://pickrate.io/blog/agent-readiness-barely-predicts-pick-rate.md): We ran Cloudflare's Agent Readiness score against every tool on the Pickrate leaderboards, then joined it to measured Pick Rate. The correlation is 0.19. Stripe scores 4 out of 16 and wins 80% of payments trials. Telnyx scores 14 out of 16 and wins nothing.
- [Why Your Analytics Can’t See AI Agents](https://pickrate.io/blog/why-your-analytics-cant-see-ai-agents.md): Google Analytics, Plausible, Fathom, Vercel Analytics — all of them are blind to AI agent traffic, and not by accident. Agents don’t run JavaScript, so the tag never fires. Here’s the mechanism, what you’re missing, and how to actually see it.
- [Your SaaS Has a Billboard. The Agent Can’t Read It.](https://pickrate.io/blog/your-saas-has-a-billboard-the-agent-cant-read-it.md): Our first SaaS leaderboard — work management — finds the household names nearly invisible to AI agents. Linear leads, Jira is named-but-not-picked, and the tools with the biggest marketing barely register. As SaaS goes headless, the selection criteria flip.
- [An Agent Sent You a Customer. Your Analytics Called It “Direct.”](https://pickrate.io/blog/agent-attribution-your-analytics-calls-it-direct.md): AI agents are quietly driving signups your analytics files under “direct.” Why agent attribution is structurally hard, which parts are actually measurable, and why anyone promising perfect AI attribution is overselling.
- [We Measure Whether Agents Pick Tools. Now They Can Pick Pickrate.](https://pickrate.io/blog/pickrate-is-webmcp-native.md): Pickrate is now WebMCP-native: open it in Chrome and an in-browser AI agent can call its tools directly. Why being callable, not just recommended, is the next stage of Agent SEO.
- [Your Coding Agent Can Now Ask Pickrate Which Tool to Pick](https://pickrate.io/blog/pickrate-mcp-server.md): Pickrate is now a remote MCP server. Add one line to Cursor or Claude Desktop and your agent can query any tool's Pick Rate, mid-task. Why being callable is the whole point.
- [What Is Pick Rate? Measuring Whether AI Agents Choose Your Tool](https://pickrate.io/blog/what-is-pick-rate.md): Pick Rate is the share of the time an AI agent picks your tool over competitors on a real task. Here's what it measures, why citations aren't the same thing, and how to read it.
- [Mintlify Is Hot With Humans. AI Agents Barely Pick It.](https://pickrate.io/blog/mintlify-hot-with-humans-agents-dont-pick-it.md): We measured which documentation platforms AI agents recommend. The buzzy hosted names lose to old open-source frameworks — a textbook case of the corpus-presence gap.
- [Agent SEO: How to Get Recommended by AI Coding Agents](https://pickrate.io/blog/agent-seo-get-recommended-by-ai-coding-agents.md): Agent SEO is the discipline of getting AI coding agents to pick your tool. A practical checklist across the four stages agents move through before they choose you.
- [Why AI Agents Aren't Picking Your SDK (and How to Fix It)](https://pickrate.io/blog/why-ai-agents-arent-picking-your-sdk.md): If agents know your SDK exists but never import it, the problem is usually Discovery, not awareness. The Probe/Serve gap, llms.txt, OpenAPI, MCP, and unbranded prompts.
- [GEO vs. Agent Selection: Citations Aren't the Same as Getting Picked](https://pickrate.io/blog/geo-vs-agent-selection-citations-arent-getting-picked.md): GEO and brand-radar tools measure whether AI talks about you. Agent selection measures whether AI chooses you. Why the distinction decides what you optimize.
- [The Agent Funnel: The Four Observable Stages of Being Chosen by AI](https://pickrate.io/blog/the-agent-funnel-four-stages-of-being-chosen.md): Awareness, Discovery, Recommendation, Adoption. The four stages an AI agent moves through before it picks your tool — and how to find the one that's capping you.

## Awareness signal — corpus footprint
Each tool report shows its presence in the open pretraining corpus: Common Crawl (the raw web) and FineWeb (the filtered derivative models train on). A tool well-represented in Common Crawl but thin in FineWeb has an awareness gap at the source — crawled, but underweight in what models actually learned from. Strongest confirmed footprints (present in the FineWeb 10BT sample):
- [Bitbucket Pipelines](https://pickrate.io/leaderboard/ci-cd/bitbucket-pipelines.md): 31 in the FineWeb 10BT sample, 1,981 Common Crawl URLs
- [Netlify](https://pickrate.io/leaderboard/hosting/netlify.md): 31 in the FineWeb 10BT sample, 816 Common Crawl URLs
- [Elasticsearch](https://pickrate.io/leaderboard/search/elasticsearch.md): 25 in the FineWeb 10BT sample, 1,968 Common Crawl URLs
- [Cloudflare](https://pickrate.io/leaderboard/cdn/cloudflare.md): 18 in the FineWeb 10BT sample, 1,899 Common Crawl URLs
- [Twilio](https://pickrate.io/leaderboard/sms/twilio.md): 17 in the FineWeb 10BT sample, 1,965 Common Crawl URLs
- [GitLab CI/CD](https://pickrate.io/leaderboard/ci-cd/gitlab-ci.md): 14 in the FineWeb 10BT sample, 1,600 Common Crawl URLs
- [New Relic](https://pickrate.io/leaderboard/observability/newrelic.md): 12 in the FineWeb 10BT sample, 1,942 Common Crawl URLs
- [Auth0](https://pickrate.io/leaderboard/auth/auth0.md): 9 in the FineWeb 10BT sample, 1,811 Common Crawl URLs
