slite-mcp-server

Enables LLM clients to fetch documents and pages from Slite workspaces through the Model Context Protocol (MCP) standard interface. Implements MCP resource handlers that translate client requests into Slite API calls, managing authentication via API tokens and returning structured document metadata and content. The server acts as a bridge between LLM applications and Slite's REST API, abstracting authentication and protocol translation.

Registers MCP resource handlers that define how LLM clients can request Slite documents through the MCP protocol. Uses the MCP SDK's resource registration API to expose Slite documents as queryable resources with URI schemes (e.g., 'slite://document/{id}'), managing resource metadata and implementing read handlers that fetch content on-demand. This enables clients to discover available resources and request them using standard MCP semantics.

Manages Slite API authentication by accepting and validating API tokens, implementing token-based request signing for all Slite API calls. The server stores the token securely (in environment variables or configuration) and injects it into HTTP headers for each API request to Slite, handling authentication errors and token expiration gracefully. Implements retry logic for transient auth failures and provides clear error messages when tokens are invalid or revoked.

Implements an HTTP client that wraps Slite REST API calls with standardized error handling, retry logic for transient failures, and timeout management. Uses exponential backoff for rate-limit and temporary errors, maps Slite API error codes to meaningful messages, and implements circuit-breaker patterns for cascading failures. Handles network timeouts, malformed responses, and API version compatibility issues transparently.

Defines MCP tools that expose Slite search functionality to LLM clients, implementing tool schemas that specify search parameters (query, filters, limit) and tool handlers that execute searches against Slite. Uses MCP SDK's tool registration API to make search discoverable and callable by LLM clients, translating tool invocations into Slite API search requests and returning ranked results. Implements result formatting for LLM consumption (summaries, snippets, relevance scores).

Implements the MCP server lifecycle using the MCP SDK's server class, managing initialization, request/response handling, and graceful shutdown. Uses stdio-based transport (stdin/stdout) to communicate with MCP clients, implementing the MCP protocol framing and message serialization. Handles server startup configuration, capability advertisement (which tools and resources are available), and error propagation back to clients through MCP error messages.

Parses Slite document responses (which may contain rich formatting, embedded media, or structured data) and formats them into text suitable for LLM consumption. Converts Slite's internal document format (likely JSON with nested content blocks) into plain text or Markdown, strips or describes media elements (images, videos), and handles special formatting (tables, code blocks, lists). Implements content truncation for very large documents to fit within LLM context windows.