@glintbase/mcp

Zero-config agent-readiness tools for AI coding agents — an MCP server that lets Claude Code, Cursor, Windsurf, OpenCode (and any MCP client) discover, crawl, and score how well a developer site works for AI agents.

No API keys. No .env. Your agent calls the tools and reasons over the results with its own subscription.

Install

Add to your client's MCP config (e.g. .mcp.json for Claude Code):

{
  "mcpServers": {
    "glintbase": {
      "command": "npx",
      "args": ["-y", "@glintbase/mcp"]
    }
  }
}

Cursor (~/.cursor/mcp.json), Windsurf (~/.codeium/windsurf/mcp_config.json) use the same mcpServers shape. For OpenCode use the mcp block with "type": "local".

Tools

| Tool | What it does | |------|--------------| | discover_surfaces | Find machine-readable entrypoints (llms.txt, OpenAPI, MCP, docs, …) | | check_reachability | Quick single-URL reachability + soft-404 detection | | parse_spec | Parse OpenAPI / llms.txt / MCP configs without crawling | | crawl_pages | Budgeted crawl with a priority queue (quick/deep) | | deep_crawl | Recover JS-rendered content with no API key | | build_knowledge_graph | Build a semantic graph from crawled pages | | run_journeys | Deterministic agent journey simulations (no LLM) | | score_readiness | Full pipeline → composite ARS score (auto-runs missing stages) | | get_remediation | Prioritized fixes with templates and expected score impact | | recheck_issues | Re-evaluate specific issues to verify fixes without full crawl | | discover_products | Detect multi-product architectures and documentation sub-trees | | get_skill | Fetch full step-by-step markdown optimization playbooks |

9 Bundled Skills (Prompts & Resources)

Available natively in your agent client as MCP Prompts (optimize-<skill>) and MCP Resources (skill://glintbase/<name>):

1. glintbase-agent-readiness: Master ARS 3.0 framework across all 6 pillars (Discovery, Access, Usability, Semantic, Architecture, Safety).
2. living-artifacts-architect: Specifications and recipes for high-density /llms.txt and /.well-known/ard.json.
3. agent-auth-handbook: Machine-to-machine authentication standards (WorkOS auth.md, OAuth 2.1, Bearer scopes).
4. streamable-mcp-builder: Streamable HTTP MCP server architecture for Next.js and Express.
5. webmcp-browser-integration: Client-side browser agent tools via window.modelContext and DOM tool tags.
6. token-tax-and-schema-optimizer: Schema friction reduction and context-window token bloat elimination.
7. flight-simulator-replay: Synthetic agent persona diagnostics (Claude Code, Cursor, Perplexity, Swarm).
8. zero-drift-ci-gate: GitHub Actions quality gate configuration and PR comments.
9. enterprise-agent-governance: Zero-Trust security, soft-404 barriers, mutation idempotency, and OWASP/ISO 42001 mapping.

Golden path

discover_surfaces  ->  deep_crawl  ->  score_readiness  ->  get_remediation

Run deep_crawl before score_readiness on JS-heavy sites (Next.js, Docusaurus, SPAs) — it recovers the real content from embedded framework payloads (__NEXT_DATA__, RSC flight, JSON-LD, <noscript>) and populates the session, so scoring reuses those pages instead of re-crawling a thin shell.

Optional: Firecrawl

For the hardest SPAs, set FIRECRAWL_API_KEY in the server env and pass profile: "deep". Not required — the zero-dep extractor handles the common cases.

Links

• Gateway & docs: scan.glintbase.dev/mcp
• Source & methodology: github.com/glintbase/glintscanner

License

Apache-2.0