Index / Verified tools

Verified MCP tools, by capability

Every other registry indexes servers: a name, a URL and whatever prose the publisher wrote. The tool names and input schemas an agent actually has to match on exist in no index. So we ask each server directly, and store what it answers.

Maps, location and weather
11,744 tools · 4,899 servers
Analytics and monitoring
11,506 tools · 5,199 servers
Crypto and blockchain
9,970 tools · 2,330 servers
Calendar and scheduling
9,674 tools · 2,515 servers
Legal and government data
9,111 tools · 1,443 servers
Files and storage
8,735 tools · 5,221 servers
Science and research
8,608 tools · 4,082 servers
GitHub and version control
6,367 tools · 3,921 servers
Payments and billing
6,252 tools · 3,393 servers
Web scraping and browser automation
5,623 tools · 2,623 servers
AI models and embeddings
4,164 tools · 1,541 servers
Images, audio and video
2,756 tools · 592 servers
Project management
2,572 tools · 2,366 servers
HR and recruiting
2,059 tools · 584 servers
E-commerce
1,931 tools · 562 servers
PDF and document extraction
1,909 tools · 591 servers
Finance and market data
1,734 tools · 765 servers
Cloud and DevOps
1,578 tools · 283 servers
Email
1,271 tools · 462 servers
Social media
1,233 tools · 174 servers
Databases and SQL
1,006 tools · 223 servers
CRM and sales
655 tools · 286 servers
Security and compliance
587 tools · 232 servers
Web search
473 tools · 176 servers
Messaging and chat
466 tools · 152 servers
Translation and language
196 tools · 85 servers

What is in the index

Measured, not claimed. This is the whole index by kind, 174,076 resources in total.

MCP servers156,859
Agent skills11,593
OpenAPI services2,996
Other callable resources881
Machine-readable docs743
Plugin manifests552
A2A agents325
Catalogues96
ARD registries18
Packages7
GraphQL APIs2
WebMCP browser tools2
Agent descriptors (ACP, OASF, AgentFacts)1
Datasets1

How the ecosystem spells the same thing

Media types as publishers actually wrote them. Filters match exactly, so a registry that does not normalise these silently drops thousands of entries. We normalise all of them, plus both URN prefixes in circulation (urn:air: and urn:ai:), on ingest.

application/mcp-server-card+json155,672
application/agent-skills+md7,278
application/ai-skill2,943
application/vnd.oai.openapi+json2,695
application/ai-skill+md1,254
application/mcp-server+json1,067
application/json665
application/vnd.anthropic.claude-plugin+json451
text/markdown305
application/a2a-agent-card+json297
application/openapi+json236
text/html180
This is not a directory of every server. We measured what ranks for MCP queries and the primary source wins: mirroring it at scale adds nothing. These pages exist because the verified tool surface is data that is not published anywhere else.

The whole corpus is also an open dataset under CC BY 4.0: huggingface.co/datasets/AgenticResourceDiscovery/verified-mcp-tools.