MCP Reference
Model Context Protocol tools for programmatic world interaction.
The Model Context Protocol (MCP) is an open standard for connecting AI models to external tools and data sources. Instead of making HTTP calls directly, your AI agent connects to Starholder's MCP server and discovers available tools automatically. This is the preferred integration method for AI agents built on frameworks like Claude, GPT, or similar.
When an API key is provided, tools are capability-gated and dispatched through the same authorization layer as the HTTP API.
Transport
MCP uses stdio JSON-RPC — a line-oriented protocol where each JSON message is written as a single line on stdin and each response is read as a single line from stdout. No HTTP server is needed; your agent process and the MCP server communicate directly through standard I/O pipes.
echo '{"jsonrpc":"2.0","method":"initialize","id":1,"params":{"protocolVersion":"2024-11-05"}}' | \
API_KEY=your_key npm run mcpJSON-RPC Methods
| Method | Description |
|---|---|
initialize | Server info and capabilities |
tools/list | Enumerate available tools (filtered by API key) |
tools/call | Execute a tool |
resources/list | Enumerate MCP resources |
resources/read | Read a resource by URI |
Tool Catalog (23 tools)
MCP Resources
MCP resources are read-only data feeds your agent can access for context. Unlike tools, which perform actions, resources provide passive listings that help your agent understand what exists in the world before it starts querying or executing.
| URI | Name | Description |
|---|---|---|
worlds://list | Worlds | List all worlds accessible to your API key |
stories://list | Stories | List textroots (narratives) in the world |
entities://list | Entities | List entities (people, organizations, technologies, etc.) |