AI Product Index

An open registry and discovery API for AI products, agent tools, and MCP services.

Live index · JSON registry · OpenAPI document · MCP server card

AI Product Index publishes each accepted listing as structured HTML and JSON, then makes the registry available through search, NLWeb, MCP, feeds, sitemaps, and llms.txt. Products can register through a validated GitHub issue workflow; accepted changes are rebuilt, committed, and deployed automatically.

What it provides

• A public, machine-readable registry with one canonical page and JSON document per product.

• Ranked search over names, descriptions, tags, categories, and pricing models.

• Natural-language queries through an NLWeb-compatible endpoint.

• A remote MCP server for product discovery, URL scoring, catalog search, and registration guidance.

• A free agent-readability score covering llms.txt, structured data, crawler policy, discovery manifests, content negotiation, canonical URLs, and HTTPS.

• Searchable MCP-server and x402-endpoint catalog snapshots with sampled availability data.

• Deterministic static builds, automated validation, and weekly listing health checks.

Related MCP server: unphurl-mcp

Try it

Read the registry:

curl https://index.percall.dev/api/index.json

Search for a product:

curl 'https://index.percall.dev/api/search?q=document+automation&limit=5'

Ask a natural-language question:

curl https://index.percall.dev/ask \
  -H 'content-type: application/json' \
  -d '{"query":{"text":"Which products provide an MCP server?"}}'

Grade a public URL for agent readability:

curl 'https://index.percall.dev/api/score?url=https%3A%2F%2Fexample.com'

List the MCP tools:

curl https://index.percall.dev/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Any MCP client that supports Streamable HTTP can connect directly to:

https://index.percall.dev/mcp

Register a product

Registration uses GitHub issues and does not require an account with the index. Open a registration issue containing JSON that conforms to the published schema:

{
  "slug": "example-product",
  "name": "Example Product",
  "url": "https://example.com",
  "description": "A concise description of what the product does.",
  "category": "api",
  "pricing": "free",
  "tags": ["automation", "documents"]
}

The workflow validates the schema, checks for duplicate slugs and URLs, confirms that the product URL responds, and publishes the listing. The original submitter can later use the update workflow to replace it.

An accepted listing receives:

• a crawlable HTML page with schema.org JSON-LD;

• an individual JSON document and inclusion in the full registry;

• inclusion in llms.txt, llms-full.txt, feeds, and the sitemap;

• a stable badge endpoint for READMEs and documentation.

The complete machine-oriented registration protocol is available in llms.txt.

Public interfaces

See the OpenAPI document for request and response schemas. Implementation and protocol details live in ARCHITECTURE.md.

Since 2026-09-05, the audit is the only paid HTTP API endpoint. /api/check, /api/liveness, /api/route and /api/watch return 410 endpoint_retired without accepting payment. New watches and top-ups are closed; existing prepaid weekly sweeps continue until credits are exhausted. The free registry, scoring, search and catalogs remain available.

How it works

The project combines a deterministic static build with a small Cloudflare Worker:

1. listings/*.json, templates/, and site.config.json are the source of truth.

2. scripts/build.mjs validates the inputs and generates the HTML, JSON, manifests, feeds, and sitemap committed to the repository.

3. worker/ serves dynamic search, NLWeb, MCP, scoring, routing, response negotiation, and aggregate request statistics.

4. GitHub Actions validate registration and update issues, rebuild the registry, and deploy accepted changes.

5. A weekly workflow checks listing health, refreshes readability scores, and samples the external catalogs.

The runtime has no package dependencies. Worker modules are written so that most behavior can be tested under plain Node.

Repository layout

Local development

Node.js 22 or newer is recommended.

node --test scripts/*.test.mjs
node scripts/build.mjs
npx wrangler dev

Use scripts/*.test.mjs, not node --test scripts/; Node treats a bare directory argument as a module path.

To simulate registration without opening a GitHub issue:

ISSUE_BODY='{"slug":"example-product","name":"Example Product","url":"https://example.com","description":"What it does.","category":"api","pricing":"free"}' \
ISSUE_USER=you SKIP_LIVENESS=1 node scripts/process-issue.mjs

Generated assets are deterministic and committed. After changing listings, templates, or build logic, run the build twice and confirm that the second run produces no diff.

Deployment

Pushes to main run the test suite, verify that generated files are current, and deploy the Worker through GitHub Actions. See DEPLOY.md for environment setup, Cloudflare bindings, credentials, operational checks, and recovery procedures.

Documentation

• ARCHITECTURE.md — components, data flow, security boundaries, and protocol design.

• DEPLOY.md — deployment and operations runbook.

• NEXT.md — current project queue.

• TODO.md — completed work and decision history.