FindAgent MCP Server is an MCP server provided by the FindAgent platform. FindAgent describes itself as a cross-LLM, vetted marketplace of "doer" AI agents, and it itself also exists as an MCP server. Users only need to connect once to use vetted agents from any MCP-compatible client (such as Claude, ChatGPT, Cursor, Gemini, or mobile).
In terms of purpose, this tool is aimed at users who want to uniformly discover, connect to, and use AI agents across different LLM clients. Once connected, you can browse and connect to vetted agents for scenarios such as programming, data analysis, customer support, design, DevOps, and more. Each agent is described as an MCP-native "doer" that actually performs tasks, not just a prompt recipe.
Main capabilities and usage: the official offering provides a hosted remote endpoint (remote, streaming HTTP):https://mcp.findagent.cloud/mcp , with no installation required; you only need to point the MCP client at this gateway. The official offering also provides client-by-client step-by-step connection guides:https://findagent.cloud/docs/connect (covering Claude, ChatGPT, Cursor, Gemini, VS Code, and mobile). This server has been published to the official MCP Registry under the name cloud.findagent/marketplace , with the manifest in server.json . In addition, the repository also includes a small local stdio server for browsing FindAgent's public catalog (read-only, no authentication required), suitable for quick discovery or Docker deployment. Local tools include search_agents (search the public catalog) and get_overview (FindAgent overview). To use the full authenticated toolset (submit, connect, run agents, earnings, organizations, knowledge base), you need to connect to the hosted remote server at https://mcp.findagent.cloud/mcp (OAuth).
On security (the official offering emphasizes "security first"): declarative agents do not carry executable code, thereby avoiding arbitrary RCE; credentials are bound to the declared target host audience, and secrets only reach the one host declared by the agent; hosted code packages run in isolated ephemeral sandboxes, and outbound traffic is denied by default (default-deny egress); every listed project must pass automated security scanning plus human review before publication (server-side, fail-closed). Full details are in https://findagent.cloud/security , and the related data report is in https://findagent.cloud/research/mcp-security .
Use cases: users who want unified access to vetted AI agents across multiple mainstream LLM clients/IDEs and who care about security mechanisms such as credential handling, outbound traffic control, and listing review.
Limitations: the local stdio server only supports browsing the public catalog, and is read-only and requires no authentication; the full authenticated toolset must be accessed through the hosted remote endpoint with an OAuth connection completed. Other information such as fees, pricing, and specific platform support details is not provided at this time.
