Semgrep folds its MCP security server into the main binary

What it does This repository previously housed a Model Context Protocol (MCP) server that wrapped Semgrep’s static-analysis engine for LLMs, agents, and IDEs. It exposed tools for vulnerability scanning, AST extraction, and rule-schema lookups, plus prompts for writing custom rules. Semgrep has since moved the server into its main binary and deprecated this standalone repo.

The interesting bit Instead of treating security as a post-commit CI step, the server turns Semgrep into a real-time consultant inside the agent loop—supporting stdio, SSE, and HTTP transports so it can live locally or remotely at mcp.semgrep.ai. The hosted option is explicitly experimental, but the architecture shows how traditional SAST tools can be repackaged as agent-native utilities.

Key highlights

• Exposes scans, AST generation, and rule-registry resources as first-class MCP primitives
• Supports stdio, server-sent events, and streamable HTTP for flexible deployment
• Available as a local Python package and container image, plus an experimental hosted endpoint
• Plugs into Cursor, VS Code/Copilot, Claude Desktop, Windsurf, and OpenAI Agents
• Active development has migrated to the main semgrep repository

Caveats

• This standalone repository is deprecated; updates now ship with the official semgrep binary
• The hosted mcp.semgrep.ai endpoint is experimental and explicitly warned to be unstable

Verdict Worth a look if you are wiring security scanning into an AI agent workflow, but head to the main semgrep repo for the current code. Skip it if you were hoping to fork a standalone MCP server here.