Install
Install the local application, connect the sources you need, and configure optional MCP clients.
The install flow below uses the default SQLite backend.
#Prerequisites
AgentGraph expects Python 3.12 or later and uv.
curl -LsSf https://astral.sh/uv/install.sh | sh
#Install AgentGraph
Install AgentGraph with every first-party connector:
uv tool install 'agentgraph-server[all]'
This makes the agentgraph command available in your shell.
Or install only the connector support you need:
uv tool install 'agentgraph-server'
uv tool install 'agentgraph-server[google]'
uv tool install 'agentgraph-server[slack]'
uv tool install 'agentgraph-server[discord]'
uv tool install 'agentgraph-server[rss]'
uv tool install 'agentgraph-server[web]'
#Start the server
To start the server in a terminal session, or see below for instructions on how to keep the server running in the background.
agentgraph server
#Connect sources
Run guided onboarding to set up each installed connector that provides an interactive flow:
agentgraph onboard
#Install the AgentGraph skill
The following command will install an AgentGraph skill in ~/.agents/skills and ~/.claude/skills:
agentgraph install-skill
#Install the browser extension
Install the AgentGraph Chrome Extension from the Chrome Web Store.
After the extension is installed, start agentgraph serve, then open a supported resource and keep it focused past the default three-second observation threshold.
#Optional: Connect ChatGPT Desktop or Claude Desktop
If you want to use AgentGraph from ChatGPT Desktop Work Mode or Claude Desktop instead of through your coding agent, print the local MCP setup instructions:
agentgraph mcp-config
For ChatGPT Desktop Work Mode, enter the printed executable in Command to launch and mcp-serve in Arguments. For Claude Desktop, add the printed JSON to its configuration file.
For coding-agent clients, use their native terminal configuration commands instead.
For ChatGPT's Codex client, register the local stdio server from your terminal:
codex mcp add agentgraph -- "$(which agentgraph)" mcp-serve
For Claude Code, use the equivalent command:
claude mcp add agentgraph -- "$(which agentgraph)" mcp-serve
#Run in the background
agentgraph serve needs to be long-running to provide ongoing polling and to support the viewer.
#macOS launchd
cat > ~/Library/LaunchAgents/com.agentgraph.serve.plist <<EOF
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
"http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>Label</key>
<string>com.agentgraph.serve</string>
<key>ProgramArguments</key>
<array>
<string>$(which agentgraph)</string>
<string>serve</string>
</array>
<key>RunAtLoad</key>
<true/>
<key>KeepAlive</key>
<true/>
</dict>
</plist>
EOF
launchctl load ~/Library/LaunchAgents/com.agentgraph.serve.plist
#Linux systemd
mkdir -p ~/.config/systemd/user
cat > ~/.config/systemd/user/agentgraph.service <<EOF
[Unit]
Description=AgentGraph local knowledge graph server
After=network.target
[Service]
ExecStart=$(which agentgraph) serve
Restart=on-failure
RestartSec=5
StandardOutput=append:/tmp/agentgraph-serve.log
StandardError=append:/tmp/agentgraph-serve.log
[Install]
WantedBy=default.target
EOF
systemctl --user daemon-reload
systemctl --user enable --now agentgraph