Progressive CLI Discovery
Agents start with `instructions`, narrow with `inspect`, and only run the tool they actually need.
Run your MCP servers behind `1mcp serve`, then let agents use CLI mode for progressive discovery instead of carrying the full tool surface in context.

1MCP solves two related problems at once:
The current product model is:
1mcp serve is the unified runtime.proxy is the preferred non-CLI path because it keeps project context and works with the widest set of AI clients.Use this page to choose a path. If you already know you want a working setup, go straight to Quick Start.
This page is for Codex, Claude, Cursor, and similar agent workflows. The default path is: add one real upstream server, start 1mcp serve, run cli-setup, then verify instructions -> inspect -> run.
instructions, inspect, and run? Go to CLI Mode.npm install -g @1mcp/agent
1mcp mcp add context7 -- npx -y @upstash/context7-mcp
1mcp serveConnect your agent to CLI mode:
1mcp cli-setup --codex
# or
1mcp cli-setup --claude --scope repo --repo-root .Verify the workflow:
1mcp instructions
1mcp inspect context7
1mcp inspect context7/query-docs
1mcp run context7/query-docs --args '{"libraryId":"/mongodb/docs","query":"aggregation pipeline"}'This is only the preview path. For prerequisites, success checks, and troubleshooting, continue to the Quick Start guide.
proxy, and direct HTTP MCP clients can share the same backend inventory.Best for first-time agent users who want a working setup before reading architecture or operator details.
Best for Codex, Claude, and similar agent loops that benefit from progressive discovery.
Best non-CLI path when you want project context, `.1mcprc`, template servers, and the widest client compatibility.
Best for MCP-native HTTP clients that can use streamable HTTP directly and do not need project context.