Graphlit

Graphlit MCP Server acts as a stdio-based protocol bridge that translates MCP client requests into Graphlit Knowledge API calls, enabling ingestion of content from Slack, Discord, Gmail, websites, podcasts, and document storage platforms. The server registers content ingestion tools that map to Graphlit's feed system, which creates persistent data connectors for each source. Content is automatically extracted to normalized formats (Markdown for documents/web, transcription for audio/video, preserved format for messages) and stored in a project container with configurable workflows.

Graphlit MCP Server exposes content retrieval tools that query the Graphlit Knowledge API's vector search engine, which embeds all ingested content and enables semantic similarity matching across documents, messages, web pages, and media transcriptions. Searches return ranked results with relevance scores, source metadata, and extracted text snippets. The retrieval pipeline integrates with Graphlit's RAG system, allowing LLM clients to augment prompts with contextually relevant content from the knowledge base.

Graphlit MCP Server supports short-term memory contents that store temporary user inputs and conversation context within a project. These memory contents are distinct from persistent ingested content and are designed for ephemeral context that should not be permanently indexed. The server provides tools to create and manage memory contents, enabling conversations to maintain context without polluting the permanent knowledge base.

Graphlit MCP Server exposes conversation management tools that create and maintain chat sessions with integrated RAG pipelines. Each conversation maintains message history and automatically retrieves relevant content from the knowledge base to augment LLM responses. The server handles conversation state management (storing messages, managing context windows) and coordinates with Graphlit's specification system (LLM configuration presets) to control model behavior, temperature, and token limits per conversation.

Graphlit MCP Server exposes collection management tools that enable organizing ingested content into named groups with independent metadata and access controls. Collections act as logical partitions within a project, allowing users to scope searches, conversations, and workflows to specific subsets of content. The server provides tools to create collections, add/remove content, and query collection membership, enabling fine-grained content organization without duplicating data.

Graphlit MCP Server exposes workflow management tools that define and execute processing pipelines for ingested content. Workflows are configured in the Graphlit dashboard and referenced via MCP tools; they can include extraction (entity recognition, summarization), transformation (format conversion, normalization), and enrichment (metadata tagging, classification) steps. The server allows querying workflow definitions and monitoring execution status, enabling content processing without custom code.

Graphlit MCP Server exposes project and specification management tools that configure the knowledge base container and LLM behavior. Projects are the top-level resource that contains all ingested content, feeds, collections, and conversations; specifications are LLM configuration presets (model, temperature, max tokens, system prompt) that control behavior across conversations and workflows. The server provides tools to query and update project settings and create/list specifications, enabling configuration without dashboard access.

Graphlit MCP Server exposes feed management tools that create and monitor persistent data connectors to external sources (Slack, Discord, Gmail, websites, podcasts). Feeds are configured once and continuously sync new content from their sources into the Graphlit project without manual intervention. The server provides tools to create feeds, monitor sync status, and manage feed credentials, enabling hands-off content ingestion for sources that produce continuous streams of data.

Graphlit MCP Server's ingestion pipeline automatically extracts and normalizes content from diverse formats into standardized representations. Documents (PDF, DOCX, PPTX) are converted to Markdown; web pages are extracted to Markdown; audio and video are transcribed to text; messages preserve original format with metadata. This normalization happens transparently during ingestion, enabling unified search and RAG retrieval across heterogeneous content types without client-side preprocessing.

Graphlit MCP Server implements the Model Context Protocol by exposing Graphlit resources (projects, contents, feeds, collections, conversations, workflows, specifications) as MCP resources and tools via stdio transport. The server uses the MCP SDK's McpServer class and StdioServerTransport to establish bidirectional communication with MCP clients (Cursor, Windsurf, Cline), translating MCP requests into Graphlit API calls and returning results as MCP responses. This enables IDE-native access to Graphlit capabilities without leaving the editor.

Graphlit MCP Server exposes specification management tools that define reusable LLM configuration presets (model, temperature, max tokens, system prompt, top-p, frequency penalty, etc.). Specifications are created once and referenced by conversations and workflows, enabling consistent LLM behavior across multiple interactions without hardcoding parameters. The server provides tools to create, list, and query specifications, allowing dynamic LLM configuration without code changes.