ax-kit MCP server
Commit Nov 10, 2026Improves the agent experience of websites, APIs, and SaaS apps by implementing standards like llms.txt, OpenAPI, and MCP.
Also known as breim/ax-kit
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 10, 2026
- License
- MIT
What is ax-kit MCP?
ax-kit improves the agent experience of websites, APIs, and SaaS apps by implementing standards like llms.txt, OpenAPI, and MCP.
Summary generated from the project's README
About ax-kit MCP server
Developers looking to make their websites, APIs, or SaaS applications accessible to AI agents use ax-kit to implement machine-readable standards. This tool helps bridge the gap between human-centric design and agentic requirements by implementing protocols like llms.txt, OpenAPI, and MCP. It is designed to work with coding agents such as Claude Code, Cursor, Codex, and Gemini CLI through the skills CLI.
The kit provides several specialized commands to manage the agent experience. You can use the improve command to find and fix implementation gaps, or the add command to implement specific capabilities from established playbooks. For assessment, the audit command provides a full diagnosis of a URL, while the score command delivers an AX Score from 0 to 100. The scanner itself is a standalone Node.js script that performs read-only checks without running JavaScript or crawling entire sites.
By following seven pillars—Reach, Discover, Read, Act, Authorize, Recover, and Transact—the tool ensures that agents can navigate, understand, and interact with your product without human intervention. It also includes a ci command to act as a pipeline gate based on your target score.
What you can do with it
- 1Automate agent-ready improvementsDirect your coding agent to identify and fix gaps in your application to make it ready for AI agents.
- 2Implement specific agent standardsAdd specific machine-readable files like llms.txt or MCP servers to your existing codebase.
- 3Diagnose agent accessibility issuesRun a full diagnostic report on a specific URL to see how an agent experiences your site.
- 4Evaluate API design for agentsReview a new API plan to ensure it is designed correctly for agentic interaction before writing code.
- 5Monitor agent experience in CI/CDSet up a pipeline gate that prevents deployments if the agent experience score falls too low.
Written from the project's README. Every use case is checked against a quote from it.
README
View on GitHubAX Kit
UX is for humans. AX is for agents.
AX Kit is an agent skill that improves the Agent Experience (AX) of a website, an API, a SaaS app, or a store. An AI agent that works for a person has no eyes, cannot solve a CAPTCHA, and has no human to ask. It can use only what your product gives to a machine.
AX Kit helps your coding agent build that layer. It implements the standards that agents use, and it measures the result with a scanner.
npx skills add breim/ax-kit
The skill works with Claude Code, Codex, Cursor, Gemini CLI, and the other agents that the skills CLI supports.
What it does
Ask your agent in plain words, or name a command.
| Command | What you get |
|---|---|
improve |
A loop: find the gaps, rank them, implement the fixes in your code, and verify each one. |
add <capability> |
One capability, implemented from a playbook. Example: add llms-txt, add mcp-server. |
audit <url> |
A full diagnosis with a report: the scanner, the checks that need judgment, and a cold-start agent journey. |
score <url> |
The AX Score and the top gaps, in seconds. |
design |
An AX review of an API, a site, or a feature before the code exists. |
ci |
A pipeline gate on the score, and a badge. |
Examples:
- "Make this app agent-ready." (
improve) - "Add an llms.txt and Markdown versions of the docs." (
add) - "Audit https://example.com for agents." (
audit) - "We are designing a new API. Review the plan for agents." (
design)
The seven pillars
The pillars follow the journey of an agent. A failure in an early pillar blocks the pillars after it.
| Pillar | Question | Playbooks |
|---|---|---|
| Reach | Can an agent get a response, and is the bot policy explicit? | bot-access |
| Discover | Can an agent find what exists? | llms-txt, sitemap-and-links, api-catalog, agent-discovery |
| Read | Can an agent understand the content with no browser? | markdown-for-agents, server-rendering, structured-data |
| Act | Can an agent call the product? | openapi-quality, mcp-server, webmcp, api-design |
| Authorize | Can an agent get access with no human present? | oauth-for-agents, self-serve-access |
| Recover | Can an agent correct itself after an error? | errors-and-limits |
| Transact | Can an agent pay? | agentic-commerce |
Each playbook is a file in skills/ax-kit/references/. It gives the reason, the status of the standard with its source, and a minimal correct implementation. It also has notes for common stacks, the checks that prove the result, security notes, and common mistakes.
The AX Score
The scanner gives one number from 0 to 100, a grade, and a list of gaps with the points that each fix can add.
$ node skills/ax-kit/scripts/ax.mjs scan https://example.com --profile content
AX Score 36 / 100 grade F (upper bound 36, coverage 100%)
Target https://example.com/
Profile content (declared)
Pillars Reach 11.43/20 > Discover 0/30 > Read 15/40 > Recover 10/10
Top gaps (points that each fix can add)
+10.00 discover.llms-txt [references/llms-txt.md]
+10.00 discover.sitemap [references/sitemap-and-links.md]
+ 5.71 reach.robots-txt [references/bot-access.md]
+ 5.00 discover.link-headers [references/sitemap-and-links.md]
...
The model has four rules that keep the number honest:
- One meaning. The score counts only checks that the scanner can measure, on stable standards. The same responses always give the same score.
- No free points. A check that does not apply is not in the score. A check with no result counts as 0 in the score and appears in the coverage.
- Emerging standards do not cost points. The report shows them in a Watch list.
- Two gates. A bot challenge on a plain request, or a page with no content without JavaScript, limits the score to 59.
Many problems that matter most with no human present need judgment. Examples: a CAPTCHA at signup, or an error message that does not say how to correct the request. The skill assesses these Review checks with evidence, and it reports them beside the score. It does not turn them into points.
Read scoring.md for the model and checks.md for the full catalog.
The scanner
The scanner is one Node script with no dependencies. It needs Node.js 18 or later. You can run it without an agent.
git clone https://github.com/breim/ax-kit.git
node ax-kit/skills/ax-kit/scripts/ax.mjs scan https://example.com --profile content
| Option | Purpose |
|---|---|
--profile content|api|webapp|commerce |
The type of the target. If you omit it, the scanner detects it. |
--json |
The full report, with the evidence of each check. |
--min-score N |
An exit code for a pipeline. It needs --profile. |
The scanner is polite and read-only:
- It sends a maximum of 80 requests, 2 at a time.
- It uses
GET,HEAD, andOPTIONS, and one read-onlyPOSTto find an MCP server. - It runs on
localhostand on private staging hosts, so you can scan before a release. - Each request has the user-agent
ax-kit/<version>. Two requests also have the published user-agent string of an AI agent before that name. They show how the site treats that agent. - It sends no telemetry.
Standards change
The standards for agents changed many times in the last year. Each playbook has a review date and links to its primary source.
The skill tells the agent to read the source again when the review date is more than 6 months old.
standards.md lists each standard with its status: stable, emerging, or legacy.
Limits
- The scanner reads the entry page and the well-known resources of one origin. It does not crawl the site.
- It does not run JavaScript. It detects WebMCP from the HTML only.
- It does not test flows behind a login, and it does not create accounts.
- A score is a measure of one surface. Scan the site, the documentation, and the API as separate targets.
Development
npm test
The tests use node:test and a local fixture server. They need no network.
The check catalog is generated from the registry. After a change to a check, run npm run catalog.
License
ax-kit MCP questions
What does AX Kit do?
It helps coding agents build an agent-friendly layer for products by implementing standards and measuring readiness with a scanner.
What is the AX Score?
The AX Score is a number from 0 to 100 that provides a grade and a list of gaps to help improve agent accessibility.
Does the scanner crawl entire websites?
No, the scanner only reads the entry page and well-known resources of a single origin and does not crawl the site.
Can the scanner test features behind a login?
No, the scanner does not test flows that require a login or create new accounts.