AI Toolkit

Agent Context Graph

Use this skill when the user asks whether Claude Code activity is being captured, how Agent Context Graph hooks work, or why graph facts are missing.

Model

Claude Code Plugin -> Claude Code Runtime Adapter -> Event Protocol -> Graph Connector -> Memgraph.

The plugin only installs runtime hook wiring. It must not assign graph meaning. Graph connectors decide which normalized events matter.

Checks

For first-time setup, run the plugin bootstrap once. It delegates to the CLI bootstrap and falls back to uvx if agent-context-graph is not installed yet:

./scripts/bootstrap.sh

If bootstrap says Memgraph is not reachable, tell the user to start Memgraph:

docker run --rm -p 7687:7687 memgraph/memgraph

Run the single CLI doctor first. It checks the same Python environment that the hook command uses:

agent-context-graph doctor --runtime claude-code --connector skills-graph --connector actions-graph --connector sessions-graph

If doctor is not available, use the strict hook smoke:

printf '{"hook_event_name":"Stop","session_id":"doctor"}' \
  | agent-context-graph hook run claude-code --connector skills-graph --connector actions-graph --connector sessions-graph --strict

Do not check skills_graph with system python3; agent-context-graph may be installed in an isolated uv tool or pipx environment.

Configuration

Connection settings live in ~/.config/context-graph/config.toml. Hook subprocesses read from this file (not environment variables).

If MEMGRAPH_URL, MEMGRAPH_USER, MEMGRAPH_PASSWORD, or MEMGRAPH_DATABASE environment variables are set when bootstrap runs, they are automatically written to the config file. No manual config set needed in that case.

View current config:

agent-context-graph config show

Set values:

agent-context-graph config set identity.user_id "username"
agent-context-graph config set memgraph.url "neo4j://coordinator:7687"
agent-context-graph config set memgraph.user "username"
agent-context-graph config set memgraph.password
agent-context-graph config set memgraph.database "dbname"

For HA clusters, use the neo4j:// scheme pointing to any coordinator. The driver handles routing and failover automatically:

agent-context-graph config set memgraph.url "neo4j://memgraph-coordinator-1:7687"

Supported keys: identity.user_id, memgraph.url, memgraph.user, memgraph.password, memgraph.database.

If skill usage is missing, inspect whether the session actually read or invoked a skill. Search/list results can be surfacing, not proven usage.

Source: memgraph/ai-toolkit — distributed by TomeVault.