erdos-renyi

erdos-renyi is an MCP server providing the privacy-preserving handshake between Erdos and Reyni. Clinicians register fine-grained EHR cohort exports with Reyni, and agents rewrite Erdos predictive SQL queries against tokenized schema names (tok_…) without ever accessing raw cell values or PHI. Results are staged locally in Arrow IPC format, gated by k-anonymity for agent-visible counts, and rendered in a private localhost Marimo dashboard for clinician review.

Quickstart

# Install dependencies
uv sync --extra dev

# Run test suite
uv run pytest

# Run the MCP server over stdio
uv run er-mcp

Reyni Gateway Setup

Add erdos-renyi to ~/Library/Application Support/reyni/servers.toml (Windows: %APPDATA%\reyni\servers.toml):

[[server]]
name       = "erdos"
launch     = ["uv", "tool", "run", "--python", "3.12", "--from", "erdos-renyi", "er-mcp"]
phi_access = true

During local development, point --from to the local repository path:

[[server]]
name       = "erdos"
launch     = ["uv", "tool", "run", "--python", "3.12", "--from", "/path/to/erdos-renyi", "er-mcp"]
phi_access = true

Agent-visible Erdos tools: erdos__get_schema, erdos__stage_query. After staging, call reyni__register_dashboard (Reyni gateway). launch_dashboard is gateway-internal only.

Agent Workflow

1. reyni__list_datasets: Obtain the safe_path for the tokenized twin export.
2. erdos__get_schema(safe_path): Inspect column names and data types (no row values or samples).
3. Rewrite SQL: Adapt the Erdos predictive query to target the local schema. The table name in Polars SQL is always cohort.
4. erdos__stage_query(safe_path, sql): Sanitize and execute SQL. If the match count is below the k-anonymity threshold ($k < 5$), the exact count is withheld (below_threshold), but the session is staged for the clinician.
5. reyni__register_dashboard(session_id): Reyni launches the Marimo dashboard internally, reads a clinician-only sidecar (session_{id}.launch.json), and stores the URL encrypted — the agent receives only a dashboard_ref.

Security Guarantees

• Zero cell egress: MCP tools only return column metadata and aggregate match counts.
• SQL Sanitization: Enforces read-only SELECT queries, rejects external filesystem functions (read_csv, eval), and clamps LIMIT to 1000 rows.
• Localhost Enclave: Staging files use Arrow IPC; dashboards bind to 127.0.0.1. Launch URLs are written to mode-0600 sidecars, not returned over MCP.