Start

AgentGraph documentation

Home GitHub Edit page

AgentGraph builds a local, queryable graph of your digital world across tools like Slack, Discord, Gmail, Google Docs, Sheets, Drive, and RSS/Atom feeds, so your AI agents can search, fetch, and reason over the same connected context you work from.

You can also teach AgentGraph about new tools quickly: connectors live in their own packages, follow a small interface, and are straightforward to build with a coding agent. That makes it practical to extend the graph to internal tools, niche SaaS products, or one-off data sources without forking the core system.

#Where to start

Install

Set up Python, sync dependencies, authenticate connectors, start the server, and connect the browser extension.

Quickstart

Get to a working local graph fast: onboard, serve, browse a supported page, and verify the first entities landed.

Configuration

Choose the config directory, database path, retention window, dwell threshold, and transport settings.

RSS

Add RSS and Atom feeds, import OPML subscriptions, and query feed articles as documents.

Extending

Learn how to extend AgentGraph with custom connectors for your own integrations, including the `BaseConnector` contract and example implementation.

Commands

Search, query, fetch, poll, auth, serve, and MCP. One reference surface for both terminal users and agent clients.

MCP tools

One page per MCP tool for search, traversal, fetch, download, and connector inspection from agent clients.

#What AgentGraph does

  • Connectors: turn messages, documents, spreadsheets, folders, and email threads into graph entities, people, and edges, and you can add your own.
  • Feeds: index RSS and Atom feed entries as documents, with feed pages represented as folders.
  • Browser-driven capture: watches supported URLs and triggers targeted fetches once you stay on a page long enough for it to matter.
  • Background refresh: polls to keep already-known resources current after the first visit.
  • Shared interfaces: the CLI, web viewer, and MCP server all operate on the same local graph.
  • Local storage: SQLite stores the graph on your machine by default.

#Reference

  • Extending for supported sources, custom connector development, and the BaseConnector interface reference.
  • RSS for feed setup, OPML imports, and RSS query examples.
  • Commands for the CLI and MCP surface.
  • MCP tools for the tool-by-tool agent interface reference.
  • Privacy for the local storage and browser observation model.

#Surfaces

Surface Use it for Entry point
CLI Ad hoc search, fetch, ingest, debugging, and operations. agentgraph search, agentgraph fetch, agentgraph serve
MCP ChatGPT developer mode, Claude Desktop, Claude Code, and other MCP clients. agentgraph mcp-config, agentgraph mcp-serve
Browser extension Passive indexing from supported browser tabs. Chrome Web Store
Viewer Visual graph inspection and manual exploration. http://127.0.0.1:8765/viewer

#Get help