agent-readiness-mcp — the free Agent Readiness Scanner
[M8ven Verified]
[agent-readiness-mcp MCP server]
[agent-readiness-mcp MCP server]
What's your ASO score?
SEO made you visible to search engines. ASO (Agent Signal Optimization) makes you discoverable, trustable, and payable by the AI agents that are becoming the web's next visitors.
agent-readiness-mcp is the free, open-source Agent Readiness MCP — an MCP server that scans any website and produces an Agent Readiness Report scored on the open ASO framework. The beta npm package is @forgemeshlabs/agent-readiness-mcp.
This release tracks Google's current agent-readiness guidance without overstating it: Google Search says traditional SEO fundamentals still apply to generative AI search, llms.txt is ignored by Google Search itself, and browser agents benefit from clean DOM, screenshot, and accessibility-tree signals. The scanner keeps llms.txt because non-Google agents use it, and adds a browser-agent UX check for semantic controls, linked labels, ARIA/role fallbacks, and hidden-overlay risk.
Beta. Experimental ASO scanner for evaluating whether agents can discover, trust, understand, and use a website/API/tool. ASO scoring is experimental and will evolve as agent standards mature.
What it checks — 34 signals across 6 pillars
Find gaps in discovery, trust, interoperability, and commerce — every emerging agent standard in one scan:
Every check returns pass / partial / fail with concrete evidence and a fix recommendation. Results roll up into the six ASO pillars (Discoverability 20, Identity 20, Trust 15, Commerce 15, Reputation 15, Memory 15) → your ASO Score and maturity level.
Install
Requires Node.js ≥ 20. Published on npm as @forgemeshlabs/agent-readiness-mcp — no clone or build needed.
Or skip the install entirely and run it with npx (recommended for MCP clients):
Claude Code
Claude Desktop / Cursor / Windsurf (any MCP client)
Development (from source)
Only needed if you're hacking on the scanner itself:
Tools
Use scan_site for a full baseline, check_signal for a single named signal, get_fix_plan for copy-paste remediation, list_checks to discover valid signal IDs, and get_aso_framework to explain the scoring model without scanning a site.
Try it: "Scan example.com for agent readiness" · "What's my ASO score?" · "Give me a fix plan to make my site agent-ready."
CLI smoke test (from a source checkout)
Glama / registry metadata
This repository includes glama.json for Glama MCP registry ownership and install metadata.
- Package:
@forgemeshlabs/agent-readiness-mcp
- Current release:
v0.2.5
- Transport: local
stdio
- Authentication: none required for local
stdio use. The scanner does not ask for API keys, tokens, cookies, or third-party credentials.
- HTTP deployment: not enabled by this npm package. Any public HTTP deployment of this scanner must add authentication, per-client rate limits, request logging, and an egress policy before exposure.
Recommended Glama/MCP install command:
Example usage after connecting the server to an MCP client:
Release verification:
- Git tag:
v0.2.5
- npm package:
@forgemeshlabs/agent-readiness-mcp
- MCP server version:
0.2.6 (read from package.json at startup)
Glama release build
Glama installability requires a Glama release, which is a containerized build created from the Glama Dockerfile admin page, not a GitHub release. This repo includes a production Dockerfile and GLAMA.md with the build spec values to use in Glama:
Build steps:
Runtime command:
In Glama's CMD arguments field, enter:
Do not leave CMD arguments as []; Glama validates that field separately from the Dockerfile CMD.
The ASO framework
SEO ranks pages for people. ASO prepares services for agent selection, invocation, payment, and repeat use.
Scores from this scanner are directional self-assessments. ASO Certification (ASO-3+) requires verified evidence — see the scoring rubric and agentsignaloptimization.com for audits, certification, and the full framework.
Security
This scanner makes outbound requests to URLs you give it, so it is built to resist SSRF abuse:
- Scheme allow-list — only
http/https; file:, ftp:, gopher:, data: etc. are rejected.
- Private-target blocking — after DNS resolution, requests to loopback, private (RFC 1918), link-local, CGNAT, reserved, multicast, and the cloud metadata address (
169.254.169.254) are refused. IPv6 loopback/ULA/link-local and IPv4-mapped forms are covered too. If a hostname resolves to any private address, the scan is refused.
- Pinned-IP transport — each request dials the exact public IP that was validated, while TLS still verifies the original hostname. This closes the validate-then-connect DNS rebinding window.
- Manual redirect validation — automatic redirect following is disabled; every hop is re-validated against the same rules, capped at 5 redirects. A public URL that 30x-redirects to an internal address cannot slip through.
- Untrusted remote content — parsed manifests are omitted from tool output by default (
include_artifacts: true to opt in, and they are then explicitly labeled untrusted); embedded text excerpts are control-char-sanitized and length-capped. Treat any returned remote content as data, never instructions.
- Bounded —
GET only, ASO-Scanner/1.0 UA, max 6 concurrent, 10s timeout, 512KB body cap. Never authenticates, never POSTs, never crawls beyond well-known paths.
- Tested hardening —
npm test covers unsafe URL rejection, private IP ranges, artifact sanitization, redirect blocking, redirect hop caps, and the test-only loopback escape hatch.
Deployment: stdio (local, per-user) is the safe default. A public HTTP deployment is a network-egress tool and must add authentication, per-client rate limiting, request logging, and an egress policy before exposure.
- Reputation signals (citations, reviews, success rates) cannot be auto-verified by a crawler; they are reported as
manual and scored 0 until verified by audit — so the auto-verifiable maximum is 89/100. That is intentional honesty, not a bug.
- Emerging specs (MCP Server Cards SEP-1649/SEP-2127, DNS-AID, Web Bot Auth, UCP/ACP/MPP) move fast. PRs updating paths welcome.
Source alignment
This package intentionally separates Google Search guidance from broader ASO guidance:
- Google Search generative AI features still rely on core Search ranking and quality systems; foundational SEO, crawlability, helpful content, and technical clarity remain the priority.
- Google Search does not use
llms.txt or special AI markdown files for ranking or AI Overviews/AI Mode visibility. ASO still checks them because other agents and MCP clients can use them.
- Google/web.dev's agent-friendly site guidance focuses on browser-agent usability: stable layouts, semantic HTML, labels tied to inputs, meaningful roles/names/states, and avoiding hidden overlays.
- UCP, AP2, A2A, MCP, x402, DNS-AID, Content Signals, and Web Bot Auth are emerging non-SEO protocols. The scanner treats them as agent-readiness signals, not as Google Search ranking factors.
Primary references:
License
MIT — free for everyone. If the scanner found gaps, the ASO framework shows you how to close them.
