@splicr/mcp-server

Implements the Model Context Protocol (MCP) server specification to expose a knowledge base as callable tools and resources that Claude and other MCP-compatible clients can discover and invoke. Routes read operations (queries, retrievals) to write operations (code generation, document creation) by translating MCP requests into internal knowledge-base queries and returning structured responses that clients can act upon.

Exposes callable tools to MCP clients through a schema registry that describes tool names, parameters, return types, and descriptions in JSON Schema format. When a client (like Claude) invokes a tool, the server receives the request, validates parameters against the schema, executes the corresponding handler function, and returns typed results. Supports multiple tools with independent schemas and execution contexts.

Implements MCP's resource model to expose knowledge-base content (documents, code snippets, architectural diagrams, etc.) as addressable resources identified by URIs. Clients request resources by URI, the server resolves the URI to the underlying knowledge-base item, retrieves the content, and returns it with metadata (MIME type, size, last-modified). Supports hierarchical resource organization and filtering by resource type.

Orchestrates a workflow where Claude reads from the knowledge base (via tools or resources) to understand requirements, patterns, and context, then generates code or documents that are written back to the Splicr system or exported to the user's environment. The server maintains context across multiple tool calls and resource retrievals within a single conversation, allowing Claude to synthesize information and produce coherent artifacts.

Manages the MCP server process lifecycle, including initialization, client connection acceptance, request routing, and graceful shutdown. Implements the MCP handshake protocol to negotiate capabilities with clients, maintains active client connections, queues and processes incoming requests, and handles errors or disconnections. Supports multiple concurrent clients and ensures request isolation between sessions.

Exposes search and indexing capabilities from the underlying knowledge base as MCP tools, allowing Claude to query the knowledge base using full-text search, semantic search, or structured filters. The server translates search queries into knowledge-base API calls, retrieves matching results, and returns them in a format Claude can process. Supports multiple search strategies (keyword, semantic, faceted) depending on the knowledge-base backend.