1mcp-app/agent

What it solves

1MCP provides a unified runtime to manage multiple Model Context Protocol (MCP) servers. It prevents "configuration sprawl" where every AI client needs separate wiring and authentication, and "agent sprawl" where AI agents are overwhelmed by too many tools and schemas in their context window.

How it works

The system runs as an aggregated runtime via 1mcp serve. It manages static servers (loaded at startup) and template servers (created based on client or session context). It offers three primary ways for agents to interact with it:

1. CLI Mode: A progressive discovery workflow where agents use instructions to understand the runtime, inspect to discover specific tools, and run to execute them.
2. Stdio Proxy: A compatibility layer that allows stdio-based AI clients to maintain project context and use template servers.
3. Direct HTTP: A streamable HTTP path for native MCP clients that do not require project context.

Who it’s for

Developers using tool-using AI agents like Claude, Cursor, and Codex who want to centralize their MCP server management and reduce the amount of noise in an agent's context.

Highlights

• Unified Runtime: Aggregates multiple MCP servers into a single process.
• Progressive Discovery: CLI mode allows agents to find and use tools only as needed rather than loading everything at once.
• lazy loading: Reduces initial schema payloads for a more stable tool surface.
• Template Servers: Enables the creation of servers based on specific per-client or per-session context.
• Context Preservation: Uses .1mcprc to keep project-specific settings across different clients.