Connect Claude Code, Codex or another agent
Agents act as you through the same capabilities as the app, under the same access rules.
Sign in once
Enter the email address of an existing account, then the code from the email and, if two-factor is enrolled, your authenticator code. Only a revocable refresh token is stored, in a file readable by your user alone. The bridge never creates an account.
Run uv run --frozen hb agent logout to revoke the session on the server and delete the file.
uv run --frozen hb agent login --api https://api.example.orgAdd it to your agent
hb agent mcp is a stdio MCP server for Claude Code, Codex or any MCP host. In Codex, add the same command and arguments under [mcp_servers.limbstead] in ~/.codex/config.toml.
The bridge adds no tools and no privilege. Every call is verified as you, and the database’s access rules decide what each tool reads or changes.
{
"mcpServers": {
"limbstead": {
"command": "/ABSOLUTE/PATH/HuggingBody/.venv/bin/hb",
"args": ["agent", "mcp"]
}
}
}What an agent can do
The in-app Assistant uses the same capabilities. It reads directly but only proposes changes, such as a complete dashboard, which you confirm in the app. Actions that need patient or staff identifiers, return credentials or publish code stay out of the Assistant.
| Goal | Calls |
|---|---|
| Build a dashboard | list_datasets → create_dashboard → get_dashboard → publish_dashboard |
| Run an analysis | list_modules → create_module / create_pipeline → run_pipeline → get_run |
| Find metrics | list_metrics, and list_modules for metrics your organization published |
| Find coded dashboards | list_coded_dashboards |
Limits
The agent has your full in-app scope. Scoped, revocable grants for unattended agents (OAuth for MCP) are not implemented yet. This setup has not yet been verified in a live Claude Code or Codex session.