Coding agents have sped up how fast code gets written, and the security findings came with it. Development and security teams can't triage and fix at that pace, so the vulnerability backlog grows. The Semgrep MCP server lets your agent take that work on: triaging findings, generating fixes, and filing tickets across your organization, so the security backlog can shrink instead of grow.
What Is the Semgrep MCP Server?
The Semgrep Model Context Protocol (MCP) server connects your AI agent to Semgrep AppSec Platform. Your agent can already write security fixes, but it can’t see which finding matters most, why last night's scan failed, or what your team decided about the last bug like this one. The Semgrep Model Context Protocol (MCP) server gives you that context. Connect Claude Code, Codex, Cursor, VS Code, ChatGPT, or another MCP client to Semgrep AppSec Platform, and your agent can detect, triage, and fix findings, debug failed scans, and manage policies without sending you to a dashboard. It's available now in beta.
Findings don’t get fixed in a dashboard, they get fixed in an editor, a terminal, or a pull request, where your agent already works. But the context that makes a fix correct lives somewhere else: which findings are open, what your team already triaged and why, which scan failed and what the log says. Today that means copying a rule ID into a search box or pasting a finding into a prompt and hoping the agent has enough to go on.
The MCP server reduces how often you have to hunt through user interfaces outside of your development environment, enabling your agent to ask the platform directly and get fast answers back or take action with Semgrep’s security tools. Every tool call goes through the same API route the platform uses, with your credential, so your agent can do what your role allows.
We’re treating the MCP server as a core way to use Semgrep. It calls the same API routes the platform uses, and new capabilities in Semgrep AppSec Platform will be available through it. We’re still investing in the UI as a place where you review findings and share results with your team.
Why Use Semgrep MCP With Your AI Agent?
Your coding agent is your software factory. It turns developer intent into working code. Semgrep is the security factory that collaborates with your software factory to reduce risk, and the MCP server is how the two work together.
A security factory stops attackers from exploiting your application in four steps: deploy, detect, remediate, and prevent. Most security tools find something and hand an issue to a person. Through the Semgrep MCP server, your agent can work on all four steps.
It can see which projects Semgrep covers and retry the scans that failed. It can start scans and search the findings. It can triage in bulk, generate fixes, and open tickets for anything that needs a person. And it can feed what it learns back in. Your agent can reuse your team’s triage memories, so a past decision shapes the next one, and it can configure Guardian rules, which check code as your coding agent writes it, so you can turn past issues into checks that run earlier.
Agentic Workflows cover the repeatable parts. Your agent can trigger and schedule them, read job status, and track credit usage, so recurring security work runs without anyone retyping the prompt.
Semgrep MCP works with whichever agent and model you run. You choose the model, so you determine what each task costs on your side, and your findings, policies, and triage context carry over unchanged if your team switches agents. Scans and workflows that Semgrep runs for you use Semgrep credits.
What Can Your Agent Do With Semgrep Findings, Scans, and Policies?
Here are five examples of things you can ask your agent to do:
Burn down a backlog. “Start with the oldest high-severity findings in this repo. Triage the ones that aren’t real, generate fixes for the rest, and file tickets for anything that needs a person.” Your agent searches the findings, triages false positives in bulk, generates fixes for the findings Semgrep can fix, and creates Jira or Linear tickets for what’s left. You review each fix before it merges, then ask for the backlog by severity to see what moved.
Check a branch before it merges. “Scan this branch against main and summarize anything new.” Your agent starts an AI-powered detection scan on the changes, compares them to the base branch, and gives you a summary of what it finds.
Find out what a new advisory touches. “A new advisory just landed. Which of our repos does it affect?” Your agent looks up the advisory, lists the dependencies involved, and returns the affected repositories.
Preview a policy change. “Dry-run this policy change before we apply it.” Your agent runs the dry run first, so you can see the effect before anything applies.
Work out why a scan failed. “Why did last night's scan fail, and can you retry it?” Your agent finds the failed scan, reads the log and error, and retries it.
These are just a few examples of areas of the platform your agent can access. The server exposes ten areas of the platform to your agent, scoped to your role or token scope and your organization's plan.
Capability | What your agent can do |
|---|
Findings and triage | Search and filter findings from Semgrep Code, Supply Chain, Secrets, and Agentic Workflows. Count by severity, project, or rule. Triage, ignore, and reopen findings in bulk. Read code snippets and finding history. |
Fixes and remediation | Generate Autofix suggestions, start AI fix jobs, and reuse your organization's triage memories. |
Scans | Start managed scans, find failed scans, read scan logs and errors, and retry scans. |
Projects | List projects, and look up a project or branch. |
Supply chain | Look up advisories, list dependencies, and find every repository affected by an advisory. |
Rules and policies | Find rules from the Semgrep Registry, test patterns against code, dry-run and apply policies, manage exceptions, and configure Semgrep Guardian rules. |
Reports | Pull backlog by severity, product, or rule, median open age, findings activity, and guardrails adoption. |
Agentic Workflows | Trigger and schedule workflows, read job status and task logs, and track credit usage. Only admins can trigger or schedule workflows. |
Automations and integrations | Create and link Jira and Linear tickets, build automations, view Slack notification settings, and check and sync source code manager (SCM) connections. |
Settings | View and update your organization’s product settings, and view a project’s Managed Scans settings. |
The server also includes a discover tool your agent can call to find the right tool for a task, and a semgrep_account tool that returns the organization you're connected to.
Semgrep MCP Server and Permissions
An agent that works on its own needs limits, and the server sets them with the access you already have. Your agent calls the same API routes the platform uses, with your credential, and each route checks your permissions as it does for any other request. If you can’t do something in the platform, your agent can't do it through the server.
You can authenticate with OAuth or with an API token. With OAuth, you sign in through your browser and pick an organization, and your agent gets your role in that organization. With a token, its scope decides which tools work, capped by the role of the token’s owner. A Web API token allows all read and write operations. A Read-only token allows reads only, so it can’t open Autofix pull requests. An Agent (CI) token covers read and write operations on scans. Member tokens aren’t supported.
Connect Your Agent to the Semgrep MCP Server
You need a Semgrep AppSec Platform account in an organization on a Teams or Enterprise plan (trials count), and an MCP client that supports remote servers over HTTP. Claude Desktop and web app, Claude Code, Chat GPT, Codex, Cursor, and VS Code all work.
The multi-tenant server URL is https://semgrep.dev/api/mcp. Single-tenant deployments use https://YOUR_TENANT.semgrep.dev/api/mcp. In ChatGPT and Codex.
Try the Semgrep MCP Server
Your findings, triage history, and policies stay in Semgrep AppSec Platform, and your agent can now work with them from the editor or terminal where you're already making the fix. Connect your agent with the setup guide.
