Driflyte

Exposes a Model Context Protocol (MCP) server that allows AI assistants to query a pre-indexed knowledge base of recursively crawled web pages. The system maintains topic-specific indexes built from web crawls, enabling assistants to retrieve contextually relevant information without making direct HTTP requests. Integration happens through MCP's standardized tool-calling interface, allowing any MCP-compatible client (Claude, custom agents) to invoke knowledge queries as native function calls.

Manages the backend crawling and indexing pipeline that discovers, fetches, and indexes web pages recursively from seed URLs. The system builds topic-specific knowledge indexes by following links within a domain or topic boundary, parsing page content, and storing indexed data for later retrieval. This is exposed to users through the Driflyte console (console.driflyte.com) and accessed by the MCP server as a pre-computed knowledge source.

Registers knowledge retrieval operations as MCP tools with standardized schemas, enabling AI assistants to discover and invoke them through the MCP protocol. The server defines tool schemas (input parameters, output format) that conform to MCP's function-calling specification, allowing clients like Claude to understand what queries are available and call them with proper type validation. This abstraction decouples the assistant from direct knowledge base access, routing all queries through the MCP interface.

Manages separate, isolated knowledge indexes for different topics or domains, allowing users to maintain multiple topic-specific knowledge bases within a single Driflyte account. Queries are scoped to a specific topic index, ensuring that knowledge from one domain doesn't contaminate results from another. This isolation is enforced at the indexing and retrieval layers, with topic identifiers passed through MCP tool parameters.

Provides a standardized MCP server interface that integrates seamlessly with Claude and other MCP-aware AI assistants. The server implements MCP's resource and tool protocols, exposing knowledge retrieval as callable functions that assistants can invoke during reasoning and response generation. Integration is bidirectional: the assistant discovers available tools at connection time and can invoke them with natural language intent, while the server returns structured results that the assistant incorporates into its context.