CodeHealth MCP Server

What it does

The CodeHealth MCP Server is a local Model Context Protocol service that gives AI coding assistants and agents access to CodeScene's CodeHealth analysis during development. It reads the source code in your local repository and returns maintainability signals, such as review findings and pre-commit safeguard results, that an agent can act on.

People typically install the server on their machine, connect it to an MCP-compatible assistant such as GitHub Copilot, Cursor, Windsurf or Claude Code, and add an AGENTS.md file to the repository. That file defines how agents should call the tools in sequence, following a review, plan, refactor, re-measure loop guided by the Code Health score. The vendor states that analysis runs locally and no source code is sent to cloud providers or LLM vendors.

It suits developers and engineering teams who already use an MCP-capable AI coding tool and want a deterministic quality check on AI-generated changes. Setup requires a CodeScene access token in the CS_ACCESS_TOKEN environment variable, and the vendor recommends frontier models for best results. A free trial is offered before a paid plan applies, and the server supports 30+ programming languages.

What it can do

• Runs CodeHealth code_health_review on a file to surface maintainability issues
• Runs pre_commit_code_health_safeguard to check changes before commit
• Feeds structured feedback to an AI agent when Code Health risk increases so it can retry
• Supports a review, plan, refactor, re-measure workflow driven by the Code Health score
• Lets teams define agent workflows through an AGENTS.md file in the repository
• Covers 30+ programming languages

Works with

• GitHub Copilot
• Cursor
• Windsurf
• Claude Code
• ChatGPT
• Codeium
• Visual Studio Code
• JetBrains
• Visual Studio

Before you choose

• Results depend on the model: the vendor strongly recommends a frontier model when the assistant offers a selector, noting older models often struggle with MCP constraints.
• The vendor's target is a Code Health score of at least 9.5 (ideally 10) for AI-ready code, and it says code below that can carry higher defect risk; these are vendor claims.
• Standalone MCP pricing differs from CodeScene team bundles. The product FAQ lists €8 or US $9 monthly and an annual discount; its card shows €86 billed yearly. Confirm the license unit and trial duration, which is inconsistent between 14 and 30 days.

Listed under

• Coding
• MCP and agent tools

Get started

Sign up for the trial, then set CS_ACCESS_TOKEN and install the server.

1. Set your token as the CS_ACCESS_TOKEN environment variable so the server can access CodeHealth analysis.
2. Install the MCP Server as an executable via Homebrew (Mac/Linux), the Windows installer, or manual download.
3. Connect the server to your AI assistant following the setup instructions for your environment, then add the AGENTS.md file to your repository.

Standalone MCP FAQ: €8/month outside the US, $9/month in the US, with an annual discount; its card shows €86 billed yearly. Team bundles are priced per active author. The product page conflicts on a 14-day versus 30-day trial, so confirm duration. A trial is not an ongoing free tier.

Source: codescene.com/product/code-health-mcp, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• Pricing | CodeScene ↗codescene.com
• Scale AI Coding Safely and Mitigate Technical Debt | CodeScene ↗codescene.com
• Fix Complex Code with AI - CodeScene ACE ↗codescene.com
• Announcement: Deterministic PR Refactoring Agents ↗codescene.com
• Developer Pack for Github Students | CodeScene ↗codescene.com
• CodeHealth™ MCP Server | CodeScene ↗codescene.com

Company website

codescene.com: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.