MCP and IDE

Your coding agent can call OLYDI directly. The MCP server exposes a governed tool surface to Claude Code, Cursor, Codex, and any MCP-compatible editor or agent, so the model works from your real findings instead of guessing at them.

Governed tools

• list_findings — the prioritized, dollar-priced findings for a repository.
• explain_risk — why a finding matters: reachability, exploit signals, and the priced exposure behind the severity.
• check_gate_state — what the gate wall and your CI currently say about a change.
• create_fix_request — open a gate-verified fix PR from a finding.

Every governed call is policy-evaluated, dollar-priced, and recorded to an immutable audit trail before it runs. A call that policy rejects fails closed: your agent sees the refusal, never a silent success.

Setup

Authenticate with the same token the CLI uses, then run the server:

npx @olydi/cli auth login
npx @olydi/cli mcp serve

Register it in your agent's MCP configuration. For Claude Code:

claude mcp add olydi -- npx @olydi/cli mcp serve

IDE

The editor surface mirrors findings and evidence chains next to the code they apply to: inline finding annotations, the live gate state of an open fix PR, and one-click fix requests routed through the same governed tools.

Scans themselves run through the CLI or the GitHub Action; the agent surface reads and acts on what they produce.