Skip to content

total-recall in Continue

Status

  • MCP support: yes (YAML config, ~/.continue/config.yaml)
  • Hook support: no
  • Session storage: ~/.continue/sessions/*.json (JSON-per-session)
  • Adapter complexity: ~430 LOC

Install the MCP server

  1. Install total-recall and uv (the example below uses uv run to launch the server straight from a checkout):
pip install total-recall    # or skip if running purely via uv
  1. Drop this into ~/.continue/config.yaml:
mcpServers:
  - name: total-recall
    command: uv
    args: [run, --directory, /path/to/total-recall, python, -m, mcp_server]

Replace /path/to/total-recall with the absolute path to your total-recall checkout (or to the installed plugin directory).

  1. Restart Continue, or reload via its settings UI.
  2. Verify: Continue's tool list should include the total-recall tools.

What you get

  • 26 MCP tools (recall, get_operator_context, check_banned, ...).
  • No hook integration — Continue exposes no lifecycle hooks.

Session ingest

total-recall autodetects Continue sessions under ~/.continue/sessions. Verify:

total-recall sources test continue

Caveats

  • YAML is indentation-sensitive; mismatched indent silently disables the server. If total-recall doesn't appear after a restart, double-check with yamllint or python -c 'import yaml,sys;yaml.safe_load(open(sys.argv[1]))' ~/.continue/config.yaml.
  • Continue's tool-call schema is closer to OpenAI's than Anthropic's; the adapter translates function_call / function_response into tool_use / tool_result blocks.
  • The source registers under the name continue (without the _dev suffix used by the Python module file).