MCP servers live across multiple registries and directories. Lists go stale, names collide, and a listing is not a safety review. This guide shows where to look, what to check, and how to let an AI agent search and install with MCPfinder.
There is no single complete catalog. Most developers combine a few sources:
MCPfinder aggregates the Official MCP Registry, Glama, and Smithery into one index your agent can query — it works with those sources, not as a replacement for your AI client.
Search by the job (filesystem, GitHub, Slack, Postgres), not by vanity rankings. Wrong tools waste context and increase risk.
Prefer clear upstream repos, multi-source listings, recent updates, and low warning-flag counts. A popular listing can still be abandoned.
Read required environment variables and OAuth scopes. Prefer least privilege; treat account-wide keys as root for that service.
Confirm transport and config shape for Claude Desktop, Cursor, Claude Code, Cline, Windsurf, or your IDE. Install JSON differs by client.
Call search_mcp_servers with a capability or keyword. Review scores and recommendation reasons across registries.
get_server_details surfaces tools, env vars, warnings, and transport before you commit.
get_install_config emits paste-ready JSON for your client. Restart and verify the tools load.
# Run MCPfinder via npx, then ask your agent to search $ npx -y @mcpfinder/server # Example ask: "find MCP servers for sqlite and show warnings"
Install MCPfinder once. Search Official Registry, Glama, and Smithery, inspect warnings, and paste client-ready config.