--- title: "Follow every agent, from first page to last call · Agentrail" description: "Agentrail recognises agents by how they behave and joins their requests across your docs, site, API and MCP server into one journey." canonical: "https://www.agentrail.co/use-cases/agent-journeys" --- ![A raccoon in a coat, sitting on a rock with a laptop covered in stickers.](https://www.agentrail.co/assets/use-cases/agent-journeys/hero-raccoon.png) Agent journeys # Follow every agent, from first page to last call. Agentrail recognises agents by how they behave and joins their requests across your docs, site, API and MCP server into one journey, and tells you how each step is known. [See How To Install](https://www.agentrail.co/docs) 2M+ agent sessions in 9 weeks \~$120K new revenue in a month at Akto The problem ## Each system logs its own piece. Nothing shares an ID. The team sees a docs hit here and an API error there, and can’t tell which agent it was, what it was trying to do, or where it gave up. **Claude Code** · 11 steps · under 4 minutes One agent, one journey - Website - Docs - MCP - Sign-up - API 1. 0:00 2. 0:44 3. 3:15 4. 3:50 - GET /llms.txt - search\_docs - agent registered - 401 · no project yet - 2½ min waiting for a person - claimed by a teammate - first project and key Without Agentrail, the same steps sit in 4 logs under 4 IDs: - visitor 8f2c - mcp\_77 - user\_4417 - key\_ab91 Watch the product ## Every other tool tries to observe the agents. You can’t: billions are already running, and more every day. So Agentrail watches the product, when agents are its users. - **Agent observability**Only sees agents you run yourself. - **Agentrail**Sits on your site, docs, API and MCP server, and sees every agent that uses them. Agent observability Watches the agents you run - Your agents - Customers’ agents - Unnamed agents Coverage**1 of 3** Agent fingerprinting ## Recognise an agent by how it behaves. It works when an agent doesn’t say who it is, and joins that agent’s steps across every interface and protocol it touches. **83%** of an agent’s steps linked Standard analytics doesn’t join them - Request headers - user-agent(none) - tlsja3 · 7c2a91 - accepttext/markdown - ordertls → h2 → get ### Request Headers, TLS and the order things arrive in. - Path, step by step - DocsGET /llms.txt - MCPsearch\_docs - Sign-upagent registered - APIPOST /v1/projects ### Behavioural What it does, step by step. - Request rhythm too steady for a person 1. 2.1s 2. 2.0s 3. 2.2s 4. 2.1s ### Timing The gaps between its requests. - Created in your product - Accountagent-test@… - API keykey\_ab91 - Projectreview-eval ### Product Accounts, keys and projects it creates. One journey, every surface ## Claude Code, eleven steps, one sign-up. **Claude Code** 11 steps · just under 4 minutes 1. 01GET /WebsiteInferred 2. 02GET /pricingWebsiteInferred 3. 03GET /llms.txtDocsInferred 4. 04GET /mcp.mdDocsInferred 5. 05MCP search\_docsMCPProven 6. 06MCP get\_pageMCPProven 7. 07GET /auth.mdDocsInferred 8. 08Agent registeredSign-upVerified 9. Hand-off: 2½ minutes waiting for a person 10. 09A teammate claimed the agentSign-upVerified 11. 10POST /v1/projectsAPIVerified 12. 11MCP create\_keyMCPVerified Example data · a sample code review product Start free ## Agent readiness checks whether agents can find and read. A free scan of your public site, docs and API, scored out of 100, with the change worth the most named under “Fix next.” Readiness checks whether agents can find and read things. A simulation checks whether they finish. [Check your site’s agent readiness](https://www.agentrail.co/agent-readiness/checker) ![A raccoon in a hooded coat, holding a magnifying glass, with a stickered laptop and a keyboard on its back.](https://www.agentrail.co/assets/use-cases/agent-journeys/banner-raccoon.png) Who it was ## An agent, a person, or unknown. People are recorded too, so you can compare them with agents. A person is identified only when they’ve signed in and your service checked the account. - **Agent**Verified or declared Verified by its registration, or declared by its user agent or MCP client. - **Person**Verified Only once they’ve signed in and your service checked the account. - **Unknown**Behaviour only Joined by behaviour until it proves who it is. ![A cartoon unmasking: an unknown visitor, revealed as an AI agent that claimed to be just a browser, standing next to Agentrail.](https://www.agentrail.co/assets/use-cases/agent-journeys/unmasked.png) Every unknown visitor gets unmasked. Faq ## Questions teams usually ask. What does the browser script record? The pages an agent opens on your site and docs, in the order they happen, so those steps can sit in the same journey as the API and MCP calls. Why do I need the edge? Many agents never run JavaScript, so a script on the page never sees them. The edge records the request itself: the docs hit, the API call, the MCP tool. How sure are the joins? Each step is marked. Inferred means the join comes from behaviour. Proven means the client named itself. Verified means your service checked a credential. What if an agent doesn’t say who it is? The steps are still joined, from the request, what it does, the gaps between calls, and what it creates in the product. The visitor stays unknown until something proves who it is. Does it track people too? It records people as well as agents, so you can compare them. A person is identified only when they’ve signed in and your service checked the account. Do I have to change my code? No. You add the edge and the browser script. Sign-up, the API and the MCP server stay as they are.