NCP (Natural Context Provider) is a tool for orchestrating all MCP servers. According to the materials, it lets AI clients face only a small number of unified tools, thereby reducing token waste, difficulty in selection, and slow responses caused by too many tools.
Purpose: To solve the "MCP paradox"—when an AI assistant faces 50+ tools, it consumes a large amount of context due to difficulty in selection, chooses the wrong tools, or frequently requests clarification. NCP enables the AI to obtain the right capabilities on demand through intelligent loading and tool discovery.
Main capabilities:
- Unified tool interface: The AI sees only 2-3 simple tools, including find (search any tool, skill, or Photon), code (directly execute TypeScript, with code mode enabled by default), and run (execute a tool separately when code mode is disabled).
- Manages 50+ tools, skills, and Photons in the background: routes requests, discovers capabilities, executes code, schedules tasks, manages health status, and caches responses.
- Code mode execution: The AI can write multi-step TypeScript workflows, combining tools, skills, and scheduling.
- Skills provide domain expertise: such as canvas design, PDF operations, document generation, etc.
- Photons allow custom TypeScript MCPs without npm publishing.
- Supports project-level configuration: each project can automatically define its own MCP.
- Internal MCPs include: Scheduler MCP (schedule tools with cron or natural language), MCP Management MCP (install and configure MCPs through natural language), Skills Management MCP (manage Anthropic Agent Skills, supporting semantic search and one-click installation), Analytics MCP (view usage statistics, token savings, and performance metrics).
- Intelligent health monitoring: automatically detects and bypasses faulty MCPs.
- Multi-configuration organization: organize MCPs by project or environment.
- Supports HTTP/SSE remote MCP, OAuth 2.1 with PKCE, and automatic wake-up of dormant servers.
- Photon runtime: For CLI installation, enablePhotonRuntime can be set in ~/.ncp/settings.json; for DXT/client builds, environment variables need to be set in the client configuration.
Usage:
- Prerequisites: Node.js 18+, npm or npx, command-line access.
- Quick start (npm): Step 1 install NCP; Step 2 optionally import existing MCPs; Step 3 configure the MCP client.
- CLI testing: ncp list shows imported MCPs; ncp find returns relevant tools; ncp config allows configuration.
- Scheduling example: ncp schedule create code:run "every day at 9am" --name "MCP Conference Scraper" --catchup-missed.
- The method for manually adding MCPs is also mentioned in the materials.
- You can browse Smithery.ai (2,200+ MCPs) or mcp.so to discover tools.
Applicable scenarios:
- When AI clients (such as Claude Desktop, Cursor, VS Code, etc.) connect to a large number of MCP tools.
- When automated execution, scheduled tasks, and cross-tool workflows are needed.
- When teams or projects need isolated MCP configurations.
- When using remote or cloud-hosted MCPs that require OAuth and dormant support.
Limitations:
- The materials do not provide specific charging or pricing information.
- No platform compatibility details are provided.
- The way to enable some features (such as the Photon runtime) varies by installation method.
- Internal MCPs are disabled by default and need to be enabled in the configuration file.
- The performance data in the materials (such as 97% token reduction and 5x response speed) does not explain the testing conditions.
The above information is entirely based on the provided materials, with no fabricated features, charges, or effects.
