aso-score-mcp is a free, open-source MCP server for calculating an "Agent Signal Optimization (ASO)" score. Its core idea is: SEO makes pages visible to search engines, while ASO makes services discoverable, trusted, understood, invoked, paid for, and reused by the AI agents that are becoming the next wave of visitors to the web. This scanner is used to assess whether a website, API, or tool has these agent-facing signals.
[Purpose] Scan any website and generate an ASO score report to help users identify gaps in agent discoverability, trust, interoperability, and commerce. The ASO score is experimental and will evolve as agent standards mature.
[Main capabilities: 34 signals, 6 pillars] Each check returns "pass / partial pass / fail," along with specific evidence and remediation suggestions. Results are aggregated into six pillars: discoverability (20 points), identity (20 points), trust (15 points), commerce (15 points), reputation (15 points), and memory (15 points), ultimately forming an ASO score and maturity level. It can also be understood as covering discovery, trust, commerce, interoperability, Google AI search guidelines, and browser agent user experience.
[MCP tools]
- scan_site: full baseline scan
- check_signal: check a single specified signal
- get_fix_plan: generate a ready-to-copy remediation plan
- list_checks: list valid signal IDs
- get_aso_framework: explain the scoring model (without scanning a site)
[Usage] Requires Node.js ≥ 20. Published to npm under the package name @forgemeshlabs/aso-score-mcp, no cloning or building required. It can be installed and used, or run directly with npx (recommended for MCP clients), and works with any MCP client such as Claude Desktop, Cursor, Windsurf, etc. Version information: current release v0.2.5, MCP server version 0.2.6 (read from package.json at startup), transport is local stdio, and local stdio requires no authentication. Example usage: "Scan example.com for ASO score" "What's my ASO score?" "Give me a fix plan to make my site agent-ready.” If developing from source, there are also corresponding commands and CLI smoke tests.
[HTTP Deployment] This npm package does not enable HTTP deployment. Any public HTTP deployment must add authentication, per-client rate limiting, request logging, and egress policies before being exposed.
[Security Design] The scanner makes outbound requests to user-supplied URLs, so built-in SSRF abuse prevention mechanisms include: only http/https protocols are allowed; after DNS resolution, loopback, private (RFC 1918), link-local, CGNAT, reserved, multicast, and cloud metadata addresses (169.254.169.254) are rejected, covering IPv6 loopback/ULA/link-local and IPv4-mapped forms; fixed IP transport is used, TLS still validates the original hostname, closing the DNS rebinding window between validation and connection; redirects are manually validated with automatic following disabled, each hop is re-validated, with a maximum of 5 redirects; untrusted remote content is by default not included in tool output (optionally enabled with include_artifacts: true and explicitly marked as untrusted), embedded text snippets are sanitized of control characters and length-limited; only GET is supported, UA is ASO-Scanner/1.0, maximum 6 concurrent, 10-second timeout, response body limit 512KB, never authenticates, never POSTs, and does not crawl beyond well-known paths. npm test covers insecure URL rejection, private IP ranges, artifact sanitization, redirect blocking, redirect hop limits, etc.
[Limitations and Notes] Reputation signals (citations, reviews, success rates) cannot be automatically verified by the crawler and are reported as "manual" and scored 0 until audit verification, so the maximum automatically verifiable score is 89/100—this is an intentional honest design. Emerging specifications (MCP Server Cards SEP-1649/SEP-2127, DNS-AID, Web Bot Auth, UCP/ACP/MPP) change rapidly; PRs updating paths are welcome.
[Source Alignment] This package deliberately distinguishes Google Search guidelines from broader ASO guidelines: Google Search's generative AI features still depend on core search ranking and quality systems, and foundational SEO, crawlability, helpful content, and technical clarity remain priorities; Google Search does not use llms.txt or special AI markdown files for ranking or AI Overviews/AI Mode visibility, but ASO still checks them because other agents and MCP clients may use them; Google/web.dev's agent-friendly site guidelines focus on browser agent usability, such as stable layout, semantic HTML, label and input association, meaningful role/name/state, and avoiding hidden overlays; UCP, AP2, A2A, MCP, x402, DNS-AID, Content Signals, and Web Bot Auth are treated as agent-readiness signals, not Google Search ranking factors. Main references:https://developers.google.com/search/docs/fundamentals/ai-optimization-guide、https://web.dev/articles/ai-agent-site-ux、https://ucp.dev。
[License and Certification] MIT licensed, free for everyone. ASO certification (ASO-3+) requires verified evidence; see the scoring rubric and the audit, certification, and complete framework at agentsignaloptimization.com.
