Model Context Protocol
Deslint MCP — the verification layer inside your AI coding agent.
Deslint ships a Model Context Protocol server that runs as a local subprocess of any MCP-compatible agent. The agent calls deterministic tools — analyze_and_fix, enforce_budget, compliance_check — before it writes code, so drift and accessibility failures never land in your PR. Zero LLM in the hot path. Zero code leaves your machine.
Install in one command
The installer detects Claude Desktop and Cursor and writes the MCP config for you. For other agents, drop the snippet below into the agent's MCP config file.
$ npx @deslint/mcp install
{
"mcpServers": {
"deslint": {
"command": "npx",
"args": ["-y", "@deslint/mcp"]
}
}
}Pick your agent
Tools the agent can call
analyze_fileReturn all design-system and a11y violations for a single file.analyze_projectScan the whole workspace and return aggregated scores.analyze_and_fixScan, apply every safe auto-fix in place, report what is left.compliance_checkProduce the full compliance report (health score, debt, WCAG) as a single structured payload.enforce_budgetGate on rule budgets from .deslintrc.json. Returns pass / fail so the agent can halt its own edits.get_rule_detailsFetch the docs, examples, and suggested fixes for any deslint rule.suggest_fix_strategyGiven a violation, return a structured fix plan the agent can execute.
Why a local MCP server — not a cloud review bot
- Pre-generation, not post-PR. The agent calls deslint before it writes the file. Drift gets stopped at the source, not 24 hours later in CI.
- Zero LLM in the hot path. Every finding is the output of a deterministic ESLint rule. Same result every run.
- Zero egress. The MCP server is a stdio subprocess. No network calls. No code leaves your machine. Safe inside enterprise firewalls and regulated environments.
Further reading: How to fix design drift in AI-generated code →