# Agent experience > Your agent experience score: can agents reach, find, read, use and pay for your product? Source: https://docs.forgeintel.co/agent-experience The agent experience check is the quickest way to see how AI agents get on with your product. Enter a domain at [forgeintel.co/agent-experience](https://www.forgeintel.co/agent-experience) and Forge visits it the way agents do: it requests your pages as Claude, ChatGPT and Perplexity, reads your agent files, finds your docs, API and MCP server, and calls your API without a key. No account is needed, and every report has its own link at `forgeintel.co/r/`. ## The report - **Your score**, out of 100, with how much it moved since the last check. - **What stops agents**: one line for each stage, with what Forge found. - **Agent traffic that can't get through**: an estimate, when the check found agents turned away or pages they can't find (see below). - **The fixes**, biggest gain first. **Improve your score** opens a free Forge account with your domain, where you get one prompt that applies every fix and can re-check with real agents. ## The five stages Each stage is worth 20 points. | Stage | The question | What Forge checks | |---|---|---| | **Reach** | Can agents get in? | Bot walls that turn AI agents away, `robots.txt` rules that block them, a missing `robots.txt`, a CAPTCHA at signup or login | | **Navigate** | Can agents find every page? | Pages agents can only reach by clicking (not in your sitemap or `llms.txt`), a sitemap, an API host with nothing at its root, whether `llms.txt` links your OpenAPI document, an API catalog, pricing or docs behind a sign-in | | **Understand** | Can agents read it cheaply? | Pages that are empty without JavaScript, `llms.txt` on your website, Markdown for agents that ask for it | | **Use** | Can agents make their first call? | Whether a `401` says where to get a key, an MCP server and a remote MCP endpoint, an SDK or CLI to install, an MCP server card, OAuth discovery for APIs and MCP servers that need sign-in | | **Pay** | Can agents pay without a person? | An agent payment rail: x402, MPP or a UCP profile | **Use** and **Pay** only count when they apply. A site with no API or MCP server skips Use, and a product with nothing for agents to pay for skips Pay. Skipped stages show as a dash and are left out of the score, so it's always measured against what your product actually offers. ## How the score works Each problem Forge finds costs points from its stage, never more than the stage's 20. | Fix | Stage | Points | |---|---|---| | Let the agents a bot wall turns away through | Reach | 15 | | Stop `robots.txt` blocking AI agents | Reach | 10 | | Let agents sign up without solving a CAPTCHA | Reach | 8 | | Add a `robots.txt` | Reach | 6 | | List the pages agents can't find in your sitemap and `llms.txt` | Navigate | 2 to 10, by the share hidden | | Add a sitemap | Navigate | 6 | | Open pricing and docs without a sign-in | Navigate | 6 | | Link your OpenAPI document from `llms.txt` | Navigate | 3 | | Publish an API catalog (`/.well-known/api-catalog`) | Navigate | 3 | | Add `robots.txt` and `llms.txt` to an API host with nothing at its root | Navigate | 2 | | Render pages without JavaScript | Understand | 10 | | Publish an `llms.txt` | Understand | 10 | | Serve Markdown when agents ask for it (`Accept: text/markdown`) | Understand | 10 | | Say where to get an API key in your `401` | Use | 10 | | Publish an MCP server (products with an API) | Use | 6 | | Add a remote MCP endpoint | Use | 6 | | Publish OAuth discovery (`/.well-known/oauth-protected-resource`) | Use | 6 | | Publish an MCP server card (`/.well-known/mcp.json`) | Use | 3 | | Show an SDK or CLI install command (products with an API) | Use | 3 | | Let agents pay without a person (x402, MPP or UCP) | Pay | 12 | - **The score** is the points kept across the stages that apply, out of 100. - **Overlapping problems are capped.** No sitemap, pages agents can't find and no `llms.txt` share one cause: agents aren't told what pages exist. Together they cost at most 18 points, shared out in proportion. - **Standards you've adopted add 2 points each**: OpenAPI, MCP, `SKILL.md`, x402, MPP, UCP, WebMCP, an API catalog, OAuth discovery and Content Signals in `robots.txt`. Bonus points never lift a product with an open fix to 100. ## Agent traffic that can't get through When the check finds agents turned away or pages they can't find, the report estimates the share of agent traffic that doesn't get through. It combines the share of agents Forge sent that were turned away with, for the rest, the share of pages they can't find. Hidden pages only count once the check has found at least five pages, so a small site isn't judged on one page. It's an estimate from the check, not your traffic. To see your real agent sessions, connect the [SDK](https://docs.forgeintel.co/sdk/overview). ## Discovery in the app Inside Forge, [Discovery](https://docs.forgeintel.co/discovery) shows the same score for each product you add, with the full fix prompt to copy, re-scans when you ask, and the paths agents take to your paid routes.