Aikido Security MCP Server

Aikido is an application-security platform aimed at engineering teams rather than security departments, and this MCP server puts one specific piece of it — the code and secrets scan — directly inside the agent that is writing the code. Instead of finding out in CI that the commit leaked a key, the agent scans, gets structured findings back, and fixes them before the commit exists.

The server is deliberately small. It exposes Aikido's scan as a tool and returns machine-readable findings, each of which the agent can triage, fix or ignore. That narrowness is the design: it adds a handful of tools to the agent's context rather than an entire security platform's API surface.

Authentication has two modes and does not require a decision up front. With nothing configured, the first tool call returns one-time sign-in links — one per Aikido region (EU, US, ME) — and you click the one matching your account; the token is then cached in the OS keychain. For CI, containers, WSL or anywhere a browser and keychain are not available, mint a Personal Access Token under Settings → Integrations → IDE → MCP and pass it as AIKIDO_API_KEY, which takes precedence over the cached browser token.

Installation is npx -y @aikidosec/mcp and the docs carry per-client configuration for Cursor, Windsurf, VS Code, Claude Code (including a bundled internal plugin), OpenAI Codex CLI, Gemini CLI, JetBrains AI / Junie, GitHub Copilot, Mistral Vibe and OpenCode, alongside pages on MCP permissions and troubleshooting. Requires Node.js 18.19.0 or newer and an Aikido account. Published under AGPL.