AgentGraph is a local-first, self-hosted CLI and MCP server that turns the sources you choose into a graph of messages, documents, people, feeds, pages, and relationships that your agent can use for reasoning.
Coding agents work well because their source of truth is already available on disk. They can search files, follow references, inspect history, and build a model of a system. AgentGraph applies that advantage to the selected digital context outside the current repository.
- Observe: when you keep a supported page focused, the Chrome extension tells the local server which resource mattered and its connector fetches it.
- Fetch: an agent or the CLI requests a specific missing or stale resource directly.
- Refresh: polling keeps known resources updated as they change.
- Expiry: content is expired using a retention model.
#Connectors
Connectors provide access to the services you use. AgentGraph provides a number of connectors for common services, but new connectors can also be added to allow integration with other services. You can extend AgentGraph by adding connectors for other online services, internal systems, niche tools, exports, local databases, and APIs.
| Connector | What it contributes |
|---|---|
| Gmail | Email threads, participants, subjects, bodies, and attachment references |
| Google Drive, Docs, Sheets | Folders, files, content, ownership, authorship, and containment |
| Slack | Channels and DMs, messages, replies, authors, mentions, and attachments |
| Discord | Channels, DMs, threads, messages, authors, mentions, and attachments |
| RSS | Feeds, posts, dates, authors, and publication relationships |
| Web | Configured pages and bookmarks with titles, text, metadata, and URLs |
Explore connectors or build your own.
#Local by design
Indexed content is stored in SQLite on your machine. Source API calls run from your machine under your credentials. The project does not operate a hosted graph service or receive indexed content through a project-controlled backend.
An MCP client you connect can read content from the local graph and is governed by that client's data practices. Read the Privacy Policy, Terms of Service, and retention model.
#Start here
Install
Install AgentGraph, connect sources, and configure an optional MCP client.
How it works
Understand observation, direct fetch, refresh, the graph model, and local data flow.
Connectors
See current coverage and how the open connector architecture expands to other services.
Retention
Learn how observations, ownership, graph connections, expiration, and bookmarks interact.
CLI reference
Search, query, fetch, observe, poll, run connector commands, and operate the local graph.
MCP tools
Connect an existing agent to search, traversal, fetch, and graph-management tools.