FindAgent
A cross-LLM marketplace of focused AI agents — including "doer" agents that execute tasks — that is itself an MCP server. Every agent passes an automated scan and a human review before it goes live.
MCP marketplace comparison
FindAgent vs other MCP marketplaces and directories
Looking for the right place to find and connect MCP servers or AI agents? Here is a factual, side-by-side look at FindAgent next to PulseMCP, Glama, Smithery, mcp.so, the AI Agents Directory, and the official MCP registry — each a real tool built for a slightly different job. We compare capabilities neutrally; every FindAgent claim is enforced in our runtime and spelled out on our security page. Facts about other products are stated as of August 2026 — check each provider’s own site for their current capabilities.
What each one is
- FindAgent
- A cross-LLM marketplace of focused AI agents — including "doer" agents that execute tasks — that is itself an MCP server. Every agent passes an automated scan and a human review before it goes live.
- PulseMCP
- A daily-updated directory that indexes MCP servers available across the internet, with news and use-case content. It points you to servers rather than hosting them.
- Glama
- A registry, in-browser inspector, and gateway for MCP servers: it indexes open-source servers, scores each with a quality-and-safety grade, and can also proxy or host them through its gateway.
- Smithery
- A registry and hosting platform for MCP servers. It can run servers as hosted remote endpoints and manages auth, credentials, and sessions for you.
- mcp.so
- A community-driven marketplace that collects and lists third-party MCP servers for discovery; servers are submitted by the community and it points you to them.
- AI Agents Directory
- A curated directory of standalone AI agents, frameworks, and tools (not specifically MCP servers) that you browse and connect to externally.
- Official MCP registry
- The official, community-supported metadata catalog of publicly accessible MCP servers, backed by MCP-ecosystem contributors. It hosts server metadata (not code) and states it assumes minimal-to-no moderation.
Capability comparison
The axes below are where these products genuinely differ. A — means a capability we could not confidently verify from that provider’s public materials — we leave it neutral rather than guess.
| Capability | FindAgent | PulseMCP | Glama | Smithery | mcp.so | AI Agents Directory | Official MCP registry |
|---|---|---|---|---|---|---|---|
| Works across many AI clientsUsable from Claude, ChatGPT, Gemini, Cursor, or any MCP-compatible client — not tied to one assistant. | Yes — Connect once, use in any MCP client | Built to work with a variety of MCP clients | One endpoint for Claude, ChatGPT, Cursor, VS Code | “Connect once, use everywhere” across MCP clients | Not verified | Not verified | Read by MCP clients and aggregators |
| Automated scan + human review before listingA listing passes both an automated security scan and a person’s review before it can go live, versus an open or auto-indexed directory. | Yes — Automated scan + human review, publish gate fails closed | No — Auto-indexed daily; no stated review or scan gate | Automated A–F quality-and-safety grade; no stated human gate | No — Self-serve with “verified” badges; no stated review gate | No — Open community submission; no stated review gate | Curated “verified” listings + open submission | No — States it assumes minimal-to-no moderation |
| Runs "doer" agents that execute tasksThe platform itself performs declared actions (a doer that calls real APIs), versus pointing you at an external server you connect to yourself. | Yes — Declarative "doer" agents call real APIs | No — Directory only — you connect to servers yourself | Lists servers; can host or proxy via its gateway | Hosts servers as remote endpoints you connect to | No — Directory only — you connect to servers yourself | No — Catalog entries link out to each standalone agent | No — Hosts server metadata, not a runtime |
| Binds your API keys to declared hostsA credential you enter is attached only to the hosts an agent declares and dropped on a redirect elsewhere. | Yes — Keys bound to declared hosts, dropped on redirect | Not verified | Managed credentials when using its gateway | Handles OAuth + injects credentials for you | Not verified | Not verified | Not verified |
| Hosted code runs in a default-deny sandboxWhen the platform runs creator code, it runs in an isolated sandbox that starts with no network access and can reach only declared hosts. | Yes — Hosted code runs isolated, no network by default | Not verified | Not verified | Not verified | Not verified | Not verified | Not verified |
See it for yourself
Every FindAgent agent shows its own security facts on its page — what code it runs, the exact hosts your keys can reach, and which tools can change data. No score, just the facts.
Reading this comparison fairly
- Every product here is a real, useful tool built for a slightly different job — a broad discovery directory, a registry, a hosting platform. This page compares capabilities, not quality, and it is not a ranking.
- These facts are stated as of August 2026. Products evolve — a directory may add review, a registry may add signals — so check each provider’s own site for its current capabilities.
- A cell marked “—” means we did not confidently verify that capability from the provider’s own public materials, not that it is absent. We leave it neutral rather than guess.
- FindAgent’s own claims are the ones we can prove: each is enforced in our runtime and review pipeline and spelled out, rule by rule, on our security page.
New to FindAgent? See how it works
01MCP marketplace comparison
FindAgent vs other MCP marketplaces
Looking for the right place to find and connect MCP servers or AI agents? Here is a factual, side-by-side look at FindAgent next to PulseMCP, Glama, Smithery, mcp.so, the AI Agents Directory, and the official MCP registry — each a real tool for a slightly different job. We compare capabilities neutrally; every FindAgent claim is enforced in our runtime and spelled out on our security page. Facts about other products are stated as of August 2026.
02What each one is
PulseMCP
A daily-updated directory that indexes MCP servers available across the internet, with news and use-case content. It points you to servers rather than hosting them.
Glama
A registry, in-browser inspector, and gateway for MCP servers: it indexes open-source servers, scores each with a quality-and-safety grade, and can also proxy or host them through its gateway.
Smithery
A registry and hosting platform for MCP servers. It can run servers as hosted remote endpoints and manages auth, credentials, and sessions for you.
mcp.so
A community-driven marketplace that collects and lists third-party MCP servers for discovery; servers are submitted by the community and it points you to them.
AI Agents Directory
A curated directory of standalone AI agents, frameworks, and tools (not specifically MCP servers) that you browse and connect to externally.
Official MCP registry
The official, community-supported metadata catalog of publicly accessible MCP servers, backed by MCP-ecosystem contributors. It hosts server metadata (not code) and states it assumes minimal-to-no moderation.
03Capability comparison
The axes below are where these products genuinely differ. A “—” means a capability we could not confidently verify from that provider’s public materials — we leave it neutral rather than guess.
| Capability | FindAgent | PulseMCP | Glama | Smithery | mcp.so | AI Agents Directory | Official MCP registry |
|---|---|---|---|---|---|---|---|
| Works across many AI clientsUsable from Claude, ChatGPT, Gemini, Cursor, or any MCP-compatible client — not tied to one assistant. | Yes — Connect once, use in any MCP client | Built to work with a variety of MCP clients | One endpoint for Claude, ChatGPT, Cursor, VS Code | “Connect once, use everywhere” across MCP clients | Not verified | Not verified | Read by MCP clients and aggregators |
| Automated scan + human review before listingA listing passes both an automated security scan and a person’s review before it can go live, versus an open or auto-indexed directory. | Yes — Automated scan + human review, publish gate fails closed | No — Auto-indexed daily; no stated review or scan gate | Automated A–F quality-and-safety grade; no stated human gate | No — Self-serve with “verified” badges; no stated review gate | No — Open community submission; no stated review gate | Curated “verified” listings + open submission | No — States it assumes minimal-to-no moderation |
| Runs "doer" agents that execute tasksThe platform itself performs declared actions (a doer that calls real APIs), versus pointing you at an external server you connect to yourself. | Yes — Declarative "doer" agents call real APIs | No — Directory only — you connect to servers yourself | Lists servers; can host or proxy via its gateway | Hosts servers as remote endpoints you connect to | No — Directory only — you connect to servers yourself | No — Catalog entries link out to each standalone agent | No — Hosts server metadata, not a runtime |
| Binds your API keys to declared hostsA credential you enter is attached only to the hosts an agent declares and dropped on a redirect elsewhere. | Yes — Keys bound to declared hosts, dropped on redirect | Not verified | Managed credentials when using its gateway | Handles OAuth + injects credentials for you | Not verified | Not verified | Not verified |
| Hosted code runs in a default-deny sandboxWhen the platform runs creator code, it runs in an isolated sandbox that starts with no network access and can reach only declared hosts. | Yes — Hosted code runs isolated, no network by default | Not verified | Not verified | Not verified | Not verified | Not verified | Not verified |
See it for yourself
Every FindAgent agent shows its own security facts on its page — what code it runs, the exact hosts your keys can reach, and which tools can change data. No score, just the facts.
Reading this comparison fairly
- Every product here is a real, useful tool built for a slightly different job — a broad discovery directory, a registry, a hosting platform. This page compares capabilities, not quality, and it is not a ranking.
- These facts are stated as of August 2026. Products evolve — a directory may add review, a registry may add signals — so check each provider’s own site for its current capabilities.
- A cell marked “—” means we did not confidently verify that capability from the provider’s own public materials, not that it is absent. We leave it neutral rather than guess.
- FindAgent’s own claims are the ones we can prove: each is enforced in our runtime and review pipeline and spelled out, rule by rule, on our security page.
New to FindAgent? See how it works
Frequently asked questions
How is FindAgent different from an MCP directory like PulseMCP or mcp.so?
A discovery directory helps you find MCP servers and then connect to each one yourself; the server is run by whoever published it. FindAgent is a marketplace that also runs the agents: most are declarative "doer" agents that ship no code and that one trusted FindAgent runtime executes, credentials you connect are bound to the hosts an agent declares, and every listing passes an automated scan plus a human review before it goes live. As of August 2026, directories are optimized for breadth of discovery, and FindAgent is optimized for connecting a vetted agent that works across your AI clients.
Do these marketplaces review listings before publishing?
It varies. Community directories and the official MCP registry are largely open or auto-indexed, and some platforms (for example Glama) publish quality and security signals for the servers they list. FindAgent requires both an automated security scan and a human review before an agent can go live, and its publish step is enforced server-side and fails closed. Check each provider’s own site for its current policy — facts here are stated as of August 2026.
Do FindAgent agents work in more than one AI assistant?
Yes. FindAgent is itself an MCP server, so an agent you connect works across Claude, ChatGPT, Gemini, Cursor, or any MCP-compatible client. You connect once and use it wherever you already work.