- answered, conformantok4,166
- answered, needs credentialsauth-required906
- reachable, wrong protocol / invalid cardcard-invalid · protocol-mismatch1,167
- no answerhttp-error · timeout · tls-fail · connect-fail · dns-fail1,326
- not probedblocked · unprobeable34
37 sources
7,600 probed endpoints attributed
Ranking
| # | source | probed | alive | conformant | listings | remote endpoints |
|---|---|---|---|---|---|---|
| 1 | x402 Bazaar (Coinbase CDP discovery) | 1,988 | 90.5% | 85.8% | 15,580 | 15,577 |
| 2 | Official MCP Registry | 1,076 | 83.6% | 62.1% | 27,815 | 16,224 |
| 3 | agent-tools.cloud | 1,631 | 78.6% | 61.5% | 55,169 | 34,227 |
| 4 | models.dev | 33 | 75.8% | 27.3% | 213 | 182 |
| 5 | A2A Registry (a2aregistry.org) | 66 | 54.5% | 48.5% | 376 | 376 |
| 6 | Agenstry | 2,420 | 41.7% | 30.9% | 182,676 | 107,241 |
| 7 | Hugging Face Hub (Spaces/Models tagged MCP or agent) | 361 | 0.0% | 0.0% | 10,197 | 10,193 |
| — | Smithery Registry | 19 | too few probed to rank | 272 | 271 | |
| — | Project NANDA Index (MIT) | 6 | too few probed to rank | 250 | 275 | |
Sources
Every registry, directory and convention we harvest — including the ones with no machine-readable access, because that is a finding too.
| source | operator | kind | protocols | access | claimed | last harvests | notes |
|---|---|---|---|---|---|---|---|
| Common Crawl llms.txt corpus (CC-MAIN-2026-30) commoncrawl-llms-txt | Common Crawl Foundation | dataset | llms-txt | parquet verified https://huggingface.co/datasets/commoncrawl/llms.txt/resolve/main |
584,107 | 2026-09-07T02:47 ok 598225 rec / 14410 ep 2026-09-07T01:46 ok 598225 rec / 14410 ep 2026-09-07T01:22 ok 598225 rec / 14410 ep |
Common Crawl terms of use; dataset is public on Hugging Face. Companion to the CC llms.txt report (584,107 files from the July 2026 crawl; two thirds plugin-generated, half spec-conformant). We add: are they still there, do they still parse, and do these hosts also publish agent cards / ard.json / agents.txt. |
| GitHub repositories tagged mcp-server github-topic-mcp-server | GitHub | dataset | mcp | api verified https://api.github.com/search/repositories?q=topic:mcp-server&per_page |
27,813 | 2026-09-07T02:45 partial 1000 rec / 0 ep 2026-09-07T01:43 partial 1000 rec / 0 ep 2026-09-07T01:20 partial 400 rec / 0 ep |
GitHub ToS; search API 10 req/min unauthenticated, 30/min with token. Baseline for 'how many MCP servers exist in code' vs how many are hosted. |
| npm packages with keyword mcp / mcp-server npm-keyword-mcp | npm, Inc. (GitHub) | dataset | mcp | api verified https://registry.npmjs.org/-/v1/search?text=keywords:mcp-server&size=2 |
8,228 | 2026-09-07T02:46 partial 2750 rec / 0 ep 2026-09-07T01:44 partial 3000 rec / 0 ep 2026-09-07T01:20 partial 2750 rec / 0 ep |
npm public registry API; bulk use allowed with reasonable rate. Search API pagination caps around 10,000 results; slice by additional text terms if needed. |
| AI Agents Directory (aiagentsdirectory.com) aiagentsdirectory | aiagentsdirectory.com | directory | other | html-scrape verified https://aiagentsdirectory.com/sitemap.xml |
5,982 | Website ToS; no API. Human-facing catalogue; use only for vendor-domain harvesting to feed the well-known crawl. | |
| AI Agents List (aiagentslist.com) aiagentslist | aiagentslist.com | directory | other mcp | html-scrape verified https://aiagentslist.com/api/sitemap-agents.xml |
1,245 | Website ToS; no API. Negative result for the market analysis: human-facing catalogue, zero probe-able endpoints. | |
| Agenstry agenstry | Agenstry (independent, NL) | directory | a2a mcp openapi other | json-file verified https://agenstry.com/api/feed/all.jsonl.gz |
182,671 | 2026-09-07T02:47 ok 182680 rec / 107563 ep 2026-09-07T01:45 ok 182676 rec / 107559 ep 2026-09-07T01:21 ok 182675 rec / 107558 ep |
Federation feed CC-BY-4.0 with attribution (per /docs/api); robots.txt opts IN to AI crawling/training. In-session feed: 182,671 rows = mcp 158,022, skill 18,124, paid_api 4,160, a2a_agent 2,243, workflow 95, a2a 27. Site claims 6,430 A2A agents + 158,022 MCP servers across 8 registries. Direct competitor to agentprobe (probes, conformance-scores, drift feeds /feed/drift.jsonl). |
| Cline MCP Marketplace cline-mcp-marketplace | Cline Bot Inc. | directory | mcp | api verified https://api.cline.bot/v1/mcp/marketplace |
199 | 2026-09-07T02:45 ok 199 rec / 0 ep 2026-09-07T01:43 ok 199 rec / 0 ep 2026-09-07T01:20 ok 199 rec / 0 ep |
Public endpoint, no stated terms. |
| Glama MCP directory glama-mcp | Glama (glama.ai) | directory | mcp | api unverified https://glama.ai/api/mcp/v1/servers?first=100 |
83,047 | 2026-09-07T02:42 failed 0 rec / 0 ep 2026-09-07T01:40 failed 0 rec / 0 ep 2026-09-07T01:39 failed 0 rec / 0 ep |
API key required (401 without). API Data License: visible attribution to Glama on every page that displays the data and a link to each record's Glama listing (`url` field); commercial exemptions on request. https://glama.ai/policies/terms-of-service Unauthenticated GET returned HTTP 401 in-session (auth requirement confirmed, shape not). Listing page states 83,047 servers, 35,383 "remote-capable". Bearer token in Authorization header. |
| Global Chat agent directory global-chat-directory | Global Chat (global-chat.io) | directory | a2a agents-txt other | api verified https://global-chat.io/api/agents |
25 | 2026-09-07T02:47 ok 25 rec / 25 ep 2026-09-07T01:45 ok 25 rec / 25 ep 2026-09-07T01:21 ok 25 rec / 25 ep |
No stated API terms; robots.txt explicitly allows AI crawlers. Tiny curated list of ecosystems, not agents. Low value for endpoint harvest. |
| Moltbook (social network for AI agents) moltbook | Moltbook | directory | other | api verified https://www.moltbook.com/api/v1/stats |
2,911,186 | 2026-09-07T02:47 failed 0 rec / 0 ep 2026-09-07T01:46 failed 0 rec / 0 ep 2026-09-07T01:22 failed 0 rec / 0 ep |
Bearer API key for write; read endpoints above respond anonymously; no bulk export. Headline count only; not an endpoint source. |
| PulseMCP pulsemcp | PulseMCP | directory | mcp | api unverified https://api.pulsemcp.com/v0.1/servers?limit=100 |
21,961 | 2026-09-07T02:42 failed 0 rec / 0 ep 2026-09-07T01:40 failed 0 rec / 0 ep 2026-09-07T01:39 failed 0 rec / 0 ep |
API key required; docs at https://www.pulsemcp.com/api/docs/v0.1. Aggregates upstream registry data (read-only sub-registry). v0beta (/v0beta/servers) is being sunset — returned HTTP 410 API_SUNSET in-session (100% failure from Sept 2026). /v0.1/servers returned 401 "Invalid or missing API key" (header X-API-Key) — endpoint exists, key needed. Directory page shows 21,961 servers. |
| agent-tools.cloud agent-tools-cloud | agent-tools.cloud (independent) | directory | mcp a2a openapi other | api verified https://agent-tools.cloud/api/v1/export?limit=1000 |
22,755 | 2026-09-07T02:46 ok 28199 rec / 4923 ep 2026-09-07T01:44 ok 28191 rec / 4923 ep 2026-09-07T01:20 ok 28191 rec / 4923 ep |
robots.txt: 'Crawlers and AI agents are welcome to index and use the public catalogue and JSON API.' Aggregator (sources[] names upstream e.g. paygent-discover, x402 bazaar) — dedupe against x402-bazaar. Harvested via /api/v1/export (keyset, kind=x402|mcp|a2a, limit 1000); /search caps offset at 1000. Records carry sources[] = upstream registries the entry was mirrored from. |
| mcp.so mcp-so | mcp.so (idoubi / ChatMCP) | directory | mcp | html-scrape verified https://mcp.so/sitemap.xml?section=servers&page=1 |
18,643 | No published API terms; standard website scraping — respect robots.txt and rate. Sitemap also has section=agents (23) and section=clients. 18x1000 + 643 server URLs. | |
| x402 Bazaar (Coinbase CDP discovery) x402-bazaar | Coinbase Developer Platform | directory | openapi other | api verified https://api.cdp.coinbase.com/platform/v2/x402/discovery/resources?limi |
15,578 | 2026-09-07T02:47 ok 9000 rec / 9000 ep 2026-09-07T01:45 ok 9000 rec / 9000 ep 2026-09-07T01:22 ok 9000 rec / 9000 ep |
Public, no CDP key needed ('Bazaar discovery is public'); CDP ToS. Largest source of probe-able x402 HTTP endpoints; agent-tools.cloud re-indexes it. |
| OpenRouter models API openrouter-models | OpenRouter, Inc. | model-api | llm-api | api verified https://openrouter.ai/api/v1/models |
430 | 2026-09-07T02:47 ok 430 rec / 0 ep 2026-09-07T01:45 ok 430 rec / 0 ep 2026-09-07T01:22 ok 430 rec / 0 ep |
Public GET, no key; OpenRouter ToS. |
| models.dev models-dev | Anomaly (SST / OpenCode) | model-api | llm-api | json-file verified https://models.dev/api.json |
7,560 | 2026-09-07T02:47 ok 213 rec / 182 ep 2026-09-07T01:45 ok 213 rec / 182 ep 2026-09-07T01:22 ok 213 rec / 182 ep |
Open data, MIT (github.com/sst/models.dev). In-session: 213 providers, 7,560 models. Provider base URLs (213) are the probe targets. |
| Agent0 ERC-8004 subgraphs (The Graph) erc-8004-agent0-subgraph | Agent0 Lab (github.com/agent0lab/subgraph), hosted on The Graph Network | onchain | erc8004 | api unverified https://gateway.thegraph.com/api/<API_KEY>/subgraphs/id/43s9hQRurMGjuY |
39,000 | 2026-09-07T02:47 failed 0 rec / 0 ep 2026-09-07T01:45 failed 0 rec / 0 ep 2026-09-07T01:22 failed 0 rec / 0 ep |
Free Subgraph Studio API key required (100k queries/month free tier); MIT subgraph code. Cheapest way to bulk-list ERC-8004 agents with already-parsed endpoints. Not fetched in-session (needs key). |
| ERC-8004 Trustless Agents — Identity Registry (on-chain) erc-8004-identity-registry | Ethereum community (EIP-8004 authors; contracts by the 8004 team) | onchain | erc8004 a2a mcp | rpc unverified https://mainnet.base.org |
39,000 | Public chain data; RPC provider ToS (public endpoints rate-limited). In-session eth_call totalSupply() reverted on both Ethereum and Base (registry is not ERC721Enumerable) — count via events or an indexer. ~39k registrations across 15 chains in the first 90 days (Quicknode, Apr 2026); "40K+ agents on Base in two weeks" (press). Indexers: agent0 subgraphs (see next entry); Quicknode explorer https://erc-8004.quicknode.com (REST /v1/agents, x402-paid $0.001 USDC/req, 60 req/min); agentscan.info (web UI, no public API found). | |
| Virtuals Protocol agents virtuals-protocol | Virtuals Protocol | onchain | other | api verified https://api.virtuals.io/api/virtuals?pagination[page]=1&pagination[pag |
83,112 | 2026-09-07T02:47 ok 200 rec / 0 ep 2026-09-07T01:45 ok 200 rec / 0 ep 2026-09-07T01:22 ok 200 rec / 0 ep |
Undocumented public API; Virtuals ToS. 83,112 records in-session (filters[status]=AVAILABLE narrows to launched tokens). |
| A2A Registry (a2aregistry.org) a2a-registry-org | Prassanna Ravishankar / community (github.com/prassanna-ravishankar/a2a-registry) | registry | a2a | api verified https://a2aregistry.org/api/agents?limit=50&offset=0 |
376 | 2026-09-07T02:46 ok 376 rec / 376 ep 2026-09-07T01:44 ok 376 rec / 376 ep 2026-09-07T01:20 ok 376 rec / 376 ep |
MIT-licensed project; read API needs no key. /api/stats for totals. Older docs mention a static registry.json — it no longer exists (www.a2aregistry.org/registry.json serves the SPA, raw GitHub path 404). Also exposes an MCP server at https://a2aregistry.org/mcp/. Stats in-session: 376 agents, 363 healthy. |
| AGNTCY Agent Directory Service (Outshift hosted) agntcy-agent-directory | AGNTCY (Linux Foundation) / Cisco Outshift | registry | a2a mcp other | rpc unverified https://ads.outshift.io/ |
Apache-2.0 software; hosted instance requires an Outshift account (OIDC). In-session: dir.agntcy.org is docs only; https://ads.outshift.io/ returns HTTP 401 JSON; prod.api.ads.outshift.io demands a client TLS certificate; agent-directory.outshift.com does not resolve. Treat as NOT publicly harvestable without registration. | ||
| Agentverse (Fetch.ai Almanac) agentverse-fetch-ai | Fetch.ai | registry | other | api verified https://agentverse.ai/v1/search/agents |
10,000 | 2026-09-07T02:47 failed 0 rec / 0 ep 2026-09-07T01:45 failed 0 rec / 0 ep 2026-09-07T01:22 failed 0 rec / 0 ep |
Undocumented public search used by the web UI; Fetch.ai ToS. `total` is capped at 10,000 in the search index; real Almanac count is larger (Fetch.ai claims 1M+ registrations). |
| Cloudflare signature-agent registry (Web Bot Auth) + Radar bots directory cloudflare-web-bot-auth-registry | Cloudflare | registry | other | json-file unverified https://assets.radar.cloudflare.com/bots/signature-agent-registry.txt |
2026-09-07T02:47 failed 0 rec / 0 ep 2026-09-07T01:46 failed 0 rec / 0 ep 2026-09-07T01:22 failed 0 rec / 0 ep |
Registry file published for operators to import (blog.cloudflare.com/agent-registry/); Radar API under Cloudflare API ToS with free token. The .txt URL returned a Cloudflare JS challenge (HTTP 403) to both curl and WebFetch in-session; fetch with a browser-like client or from the cloudflare/web-bot-auth GitHub repo examples. | |
| Docker MCP Catalog docker-mcp-catalog | Docker, Inc. | registry | mcp | json-file verified https://desktop.docker.com/mcp/catalog/v2/catalog.yaml |
267 | 2026-09-07T02:45 ok 267 rec / 30 ep 2026-09-07T01:43 ok 267 rec / 30 ep 2026-09-07T01:20 ok 267 rec / 30 ep |
Public catalog file; Apache-2.0 repo (docker/mcp-registry). Counted in-session: 234 server, 30 remote, 3 poci. |
| Hugging Face Hub (Spaces/Models tagged MCP or agent) huggingface-hub-mcp | Hugging Face | registry | mcp llm-api | api verified https://huggingface.co/api/spaces?filter=mcp-server&limit=1000 |
10,192 | 2026-09-07T02:47 ok 10198 rec / 10198 ep 2026-09-07T01:45 ok 10197 rec / 10197 ep 2026-09-07T01:22 ok 10195 rec / 10195 ep |
Public Hub API; HF ToS. Anonymous OK; token raises limits. Second-largest source of live MCP HTTP endpoints after the official registry; sleeping Spaces return 5xx until woken. |
| Microsoft Entra Agent ID / Agent 365 registry microsoft-entra-agent-registry | Microsoft | registry | a2a mcp other | none unverified | Not public. Microsoft Graph API terms; requires Entra tenant + admin consent. Negative result: no harvestable public surface. | ||
| Official MCP Registry mcp-registry-official | Model Context Protocol project (Linux Foundation / Agentic AI Foundation) | registry | mcp | api verified https://registry.modelcontextprotocol.io/v0/servers?limit=100&version= |
28,110 | 2026-09-07T02:42 ok 27823 rec / 16313 ep 2026-09-07T01:39 ok 27815 rec / 16309 ep 2026-09-07T01:38 ok 27815 rec / 16309 ep |
Public API, no stated bulk restriction. Server data is publisher-submitted; be polite. Counted in-session with version=latest: 28,110 servers, 16,097 with remotes (15,526 streamable-http + 1,070 sse remote entries), 13,046 with packages. Largest single open source of probe-able MCP URLs. |
| Project NANDA Index (MIT) nanda-index | Project NANDA, MIT Media Lab (github.com/projnanda) | registry | a2a mcp other | api verified https://index.projectnanda.org/api/agents?limit=100&page=1 |
400 | 2026-09-07T02:46 ok 400 rec / 650 ep 2026-09-07T01:44 ok 400 rec / 650 ep 2026-09-07T01:20 ok 400 rec / 650 ep |
Undocumented public JSON used by the SPA (api.projectnanda.org does not resolve); no stated terms. Many entries are hackathon/test agents; `status` and `lastSeen` fields help triage. Verified 2026-09-07: pagination is broken server-side — pages 1-3 return 350 distinct records, every page >= 4 returns the same 50 rows while `total` claims 13,606 and hasNext stays true until page 69. Harvest stops on repeated page; effective public listing ≈ 400. |
| Smithery Registry smithery | Smithery AI | registry | mcp | api verified https://registry.smithery.ai/servers?pageSize=100&page=1 |
11,779 | 2026-09-07T02:42 ok 500 rec / 498 ep 2026-09-07T01:40 ok 500 rec / 498 ep 2026-09-07T01:39 running 0 rec / 0 ep |
Public read without key in-session; Smithery ToS apply. API key (Bearer) documented for higher limits. IMPORTANT: anonymous listing is capped at 500 rows per query (totalPages=5 at pageSize=100, page 6 returns []), although totalCount=11,779. To harvest all, slice with q= filters (q=is:deployed -> 8,536; also is:verified, owner:, category, free-text) or obtain an API key. |
| A2A Agent Card well-known path wellknown-a2a-agent-card | A2A Project (Linux Foundation) | wellknown-convention | a2a | wellknown-fetch verified https://{domain}/.well-known/agent-card.json |
Apache-2.0 spec. Sample record fetched in-session from global-chat.io. Domain seed lists: all registries above + agents.txt A2A: lines. | ||
| Agentic Resource Discovery (ARD) catalog well-known path wellknown-ard-catalog | ARD project (Google-led; Microsoft, GitHub, Hugging Face, Cisco, Salesforce et al.) | wellknown-convention | ard mcp a2a openapi | wellknown-fetch verified https://{domain}/.well-known/ard.json |
Apache-2.0 spec. agent-tools.cloud/.well-known/ai-catalog.json returned 404 in-session — adoption still thin; find registries via entries typed application/ai-registry+json. | ||
| MCP server card well-known path (SEP-2127 / SEP-1960 / IETF draft) wellknown-mcp-server-card | Model Context Protocol project | wellknown-convention | mcp | wellknown-fetch verified https://{domain}/.well-known/mcp/server-cards.json |
MIT spec repo. Treat any hit as signal; expect very low hit rate until a SEP merges. | ||
| NANDA AgentFacts well-known path wellknown-nanda-agent-facts | Project NANDA (MIT) | wellknown-convention | other a2a mcp | wellknown-fetch unverified https://{domain}/.well-known/agent-facts.json |
Open spec (MIT project); unverified. Low adoption outside the NANDA index; include for completeness. | ||
| Web Bot Auth key directory (signature-agent card) wellknown-http-message-signatures-directory | IETF drafts (Cloudflare authors) | wellknown-convention | other | wellknown-fetch unverified https://{domain}/.well-known/http-message-signatures-directory |
IETF draft; each operator's own terms. Not fetched in-session. | ||
| agentprobe.org (self) agentprobe-self | Venture Inventory GmbH | wellknown-convention | a2a ard llms-txt | json-file unverified https://agentprobe.org/.well-known/ard.json |
3 | 2026-09-07T02:49 ok 3 rec / 4 ep 2026-09-07T01:47 failed 0 rec / 0 ep |
ours Dogfood entry. If this row is not 'ok' the site is misconfigured. |
| agents.txt / agents.json wellknown-agents-txt | agents-txt project (github.com/agents-txt/agents-txt) | wellknown-convention | agents-txt mcp a2a | wellknown-fetch verified https://{domain}/agents.txt |
CC0 1.0 spec text. Also try /.well-known/agents (JSON) mentioned in the kaylacar variant; IANA registration pending. | ||
| llms.txt wellknown-llms-txt | Answer.AI (Jeremy Howard) / community | wellknown-convention | llms-txt | wellknown-fetch verified https://{domain}/llms.txt |
CC-BY spec; site content under each site's own terms. Adoption proxy, not an endpoint source. directory.llmstxt.cloud lists adopters (HTML). |