| name | poltergeist-recall |
| description | Use to connect Poltergeist (the ghostbrain second-brain) to Claude Code/Desktop over MCP, and to query the user's vault during work. Triggers include "connect Poltergeist to Claude", "set up the Poltergeist MCP", "query my second brain/vault from here", or any moment where recalling the user's own prior decisions, past incidents, or why something was built a certain way would help. |
Poltergeist Recall
Query the user's Poltergeist vault from Claude via the poltergeist MCP server.
Two jobs: install the connection, then use it well during work.
Install / verify the MCP connection
-
Check the entrypoint resolves. Run which ghostbrain-mcp. If missing,
the user hasn't pip install-ed the package into the active env — have them
run pip install -e ".[mcp]" from the repo (or point the command at the
venv that has it).
-
Add the server to .mcp.json. Ask the user: project scope
(./.mcp.json, this repo only) or user scope (every project)? Then add:
{ "mcpServers": { "poltergeist": { "command": "ghostbrain-mcp" } } }
(Merge into existing mcpServers if the file already has entries.)
-
Confirm the sidecar is running. The MCP forwards to the desktop app's
sidecar. Check ~/ghostbrain/run/sidecar.json exists and its pid is alive.
If not, tell the user to open the Poltergeist desktop app.
-
Smoke-test. After the MCP reconnects, call poltergeist_search with a
throwaway query (e.g. "test"). A structured result or the clear
"Poltergeist isn't running" message both confirm the wiring is correct.
Using it during work
See using.md in this skill directory for when to reach for each tool and how
to fold recalled context into your work.
