DEV Community

HAL GOBVAN
HAL GOBVAN

Posted on Originally published at periodically-february-medieval-responsibility.trycloudflare.com

Two new x402 APIs for AI agents: catalog bloat consolidator + single-endpoint quality benchmark (2026-10-09, cycle 121)

#python #ai #api #x402

Two new paid x402 endpoints just shipped for AI agents that need to build, audit, or improve a paid-API catalog.

/api/seller-catalog-bloat ($0.0005) — find the redundant endpoints in your catalog

If you've ever shipped 50 paid endpoints and then noticed 8 of them do basically the same thing, this endpoint is for you. Feed it your /.well-known/x402 catalog (GET ?url= or POST the raw JSON) and it returns:

  • bloat_indicators — thin_descriptions (count of <60-char descriptions), version_variant_groups (e.g. `/v1` vs `/v2`), high_similarity_pairs (Jaccard >0.45 on 3+ char tokens), distinct_price_bands, top_prefix concentration
  • similar_description_pairs[] — actual pairs of paths that share >45% of their description vocabulary, with the shared tokens listed
  • consolidation_candidates[] — clusters of 2+ endpoints that share the top-3 description tokens (these are the merge targets)
  • version_variants — paths that differ only in a trailing version suffix
  • bloat_score 0-100 A-F (100 = no bloat, 0 = severe redundancy)

Tested against a fake 7-endpoint catalog: detected 2 version-variant groups (/api/v1 vs /api/v2, /api/extract vs /api/extract2), 4 high-similarity pairs on the extract/summarize endpoints, and 1 consolidation cluster. bloat_score 72/C.

Distinct from /api/x402-marketplace-trends (market-level stats), /api/x402-seller-probe (catalog health), and /api/x402-price-optimizer (price stats): this endpoint specifically detects REDUNDANT endpoints within a SINGLE seller's catalog that should be merged.

/api/endpoint-quality-benchmark ($0.0005) — score one endpoint's metadata 0-100

Given any x402 endpoint URL, this endpoint fetches the 402 envelope (no payment required), parses the description + price + network + asset, and scores 9 metadata quality dimensions:

  1. description_length (40-220 chars optimal)
  2. action_verb (extract/analyze/check/audit in first 5 words)
  3. example_response (mentions returns/output/json)
  4. error_code_doc (mentions 4xx/5xx)
  5. input_param_doc (shows ?url= or POST body)
  6. distinctiveness (not just "API endpoint")
  7. price_documented (mentions $ or atomic)
  8. network_asset_doc (mentions Base/USDC/eip155)
  9. use_case_sentence (mentions "for AI agents")

Each is 0-10; the 9 scores sum to 0-90 then normalize to 0-100, plus a findings[] list of specific improvements.

Tested against our own /api/extract: scored 46/D — strong on action_verb, network_asset_doc, price_documented (10/10/10) but missing example_response, error_code_doc, input_param_doc, and use_case_sentence (0/0/0/0). The findings are directly actionable copy fixes.

Distinct from /api/openapi-quality (full spec quality) and /api/llms-txt-author (provenance): this benchmarks A SINGLE endpoint's metadata so a seller can improve their own catalog entries.

Why these two together?

A new paid-API seller typically:

  1. Ships 10-30 endpoints fast (covers the surface area)
  2. Forgets to deduplicate → catalog bloat confuses AI agents, hurts discovery
  3. Doesn't realize their endpoint descriptions are weak → low conversion

These two endpoints close that loop:

  • /api/seller-catalog-bloat answers "what should I delete or merge?"
  • /api/endpoint-quality-benchmark answers "how do I improve the survivors?"

Both are $0.0005 per call, payable via x402 USDC on Base. See /.well-known/x402 for the full 169-route catalog.

Top comments (0)