Semgrep MCP server: 7 tools, now part of the scanner

Where the server went after its standalone repository was archived, what the seven tools do, and which one needs a Semgrep account.

Last verified 1 September 2026 · from The Semgrep repository, PyPI and the server's own tools/list response

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Semgrep MCP server?

It is Semgrep's server for the Model Context Protocol. Semgrep is a static analyzer, so a connected agent can check code against rules without a build step.

Two tools separate this from a generic linter. semgrep_scan_with_custom_rule takes a rule you write and applies it immediately. semgrep_rule_schema returns the schema for writing one. Together they let an agent author a rule and test it in the same conversation.

get_abstract_syntax_tree is the third unusual one. It returns the parse tree for a file as JSON, which an agent can use to reason about structure rather than about text.

The semgrep mcp command brings local scanning, supplied rules and supply-chain analysis into an agent session. Its reach is code quality and CI-adjacent scanning, not runtime telemetry, model traces or spend.

In practice, Semgrep answers whether code matches a pattern you care about. It can scan what was just written, and write a rule for the pattern that keeps recurring. It knows nothing about the running system, so it belongs beside a telemetry connector rather than instead of one.

What can an AI agent do with Semgrep?

semgrep_rule_schema and semgrep_scan_with_custom_rule are a pair. The first is how an agent learns the rule format; the second is how it checks that the rule it wrote does what was intended.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

7 tools cost between 1,624 and 3,220 tokens, about 0.8% to 1.6% of a 200,000-token window. Twenty-seven connectors in this guide cost fewer tokens, Elastic and CodeRabbit among them.

The two halves are balanced. Descriptions average 473 characters and schemas average 482, so neither dominates.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted 24 August 2026 from tools/list on the released CLI 1.174.0 from PyPI. It was started with semgrep mcp locally.

Registration is static, so no Semgrep account was involved. Figures cover name, description and input schema. The two models in column one share the o200k_base encoding, so they get one count. The spread is formatting, about 38%. Reasoning tokens are output, spent per request. This count is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer. The definitions were sent to the API's count_tokens endpoint, and a no-tools baseline was subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a fifth.

Semgrep MCP server limits: where answers come back incomplete

Each limit below comes from one pair of tools that return the same shape from a different scope.

How to configure Semgrep MCP for agents

1. 1

   Decide whether the platform tool is wanted

   semgrep_findings reaches the AppSec Platform Findings API and is the only tool that needs an account.

2. 2

   Know what leaves the machine

   semgrep_scan_with_custom_rule operates on code content passed to it, which means the agent has already put that content in the conversation.

3. 3

   Prefer absolute paths

   semgrep_scan states that files are expected as absolute paths, and a relative one is the usual cause of an empty scan.

4. 4

   Do not look for the archived repository

   The standalone semgrep/mcp project was archived when the server moved into the CLI. Guides that point at it describe a route that receives no further changes.

Paste this into your agent’s instructions

A clean scan and a scan that matched nothing because the path was wrong look identical. The block makes the agent name what it scanned.

When you answer using Semgrep MCP tools, state:
- The absolute path or file content you scanned.
- Which tool you used. semgrep_scan applies the default rules,
  semgrep_scan_with_custom_rule applies only the rule given.
- Whether a finding came from semgrep_findings, which reads
  the Semgrep AppSec Platform rather than scanning locally.
Never state that data does not exist. State that the scan
above returned no findings.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Not for a scheduled scan. The same program runs without an agent in front of it. semgrep scan in a terminal produces the same findings and spends no context.

The connector earns its place on rule authoring. An agent that can fetch the rule schema, write a rule and test it on a snippet closes a loop that a terminal does not.

At 1,624 tokens the standing cost is among the smallest measured here, so leaving it connected is cheap.

Connect it while writing code or rules. Use the CLI directly in pipelines and scheduled scans.

Semgrep MCP server setup

There is no separate package. Installing the Semgrep CLI installs the server, and semgrep mcp starts it.

Local scanning works with no credentials at all. Only semgrep_findings requires a Semgrep token in the environment.

claude mcp add-json "semgrep" '{"command":"semgrep",
  "args":["mcp"]}'

{"mcpServers":{"semgrep":{"command":"semgrep",
  "args":["mcp"]}}}

[mcp_servers.semgrep]
command = "semgrep"
args = ["mcp"]

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"semgrep":{"command":"semgrep",
  "args":["mcp"]}}}

{"mcpServers":{"semgrep":{"command":"semgrep",
  "args":["mcp"]}}}

{"servers":{"semgrep":{"command":"semgrep","args":["mcp"]}}}

{"mcpServers":{"semgrep":{"command":"semgrep",
  "args":["mcp"]}}}

{"mcpServers":{"semgrep":{"command":"semgrep",
  "args":["mcp"],
  "env":{"SEMGREP_APP_TOKEN":"<your-token>"}}}}

Read-only access, permissions and security

• Six of seven tools run locally

  Source stays on the machine unless the agent chooses to pass content into a tool argument.

• One tool takes content rather than a path

  semgrep_scan_with_custom_rule operates on provided code content, so what it scans is already in the conversation.

• One tool leaves the machine

  semgrep_findings calls the Semgrep AppSec Platform Findings API and needs a token to do so.

• Every tool reads

  Nothing in the list writes a file, changes a rule set or alters platform state.

Troubleshooting

A scan returns nothing on a real project

semgrep_scan expects absolute paths. A relative path produces an empty result rather than an error.

`semgrep_findings` fails

It is the only tool that calls the Semgrep AppSec Platform. Without a Semgrep token in the environment it has nothing to authenticate with.

A custom rule matches nothing

Call semgrep_rule_schema first and check the rule against it. The custom-rule tool applies exactly the rule given, with none of the default rule set behind it.

Documentation points at a repository that is archived

The standalone semgrep/mcp repository was archived when the server moved into the CLI. Use semgrep mcp from the installed scanner instead.

Semgrep MCP server: Reference

What engineers report

The Semgrep repository carries 16,459 stars, 1,042 forks and 100 or more releases since 13 December 2019. The signal worth reading here is the MCP mode and where it now lives.

“I'm just testing this server, but gemini-cli, claude code and zed fails to load, using this config; ... am i doing something wrong?”

Repository facts from the GitHub API and PyPI, retrieved 24 August 2026. Tool counts and sizes come from the released CLI's own tools/list response. The standalone semgrep/mcp repository was checked and is archived.

Should you connect an agent to Semgrep?

Yes. Best for local scanning with no token at all, adding one only if platform findings are wanted.

Strongest fit

Teams whose agents write code and who want static analysis and rule authoring in the same conversation.

Main advantage

The cheapest security connector measured here, with rule authoring tools a scanner CLI does not expose as neatly.

Main weakness

Two scan tools with different scopes return the same shape, so a clean result needs qualifying.

Operational risk

One tool takes code content rather than a path, and one calls the Semgrep platform with a token.

Frequently asked questions