check_packageValidate one package before adding it.
Typosquat, deprecation, known CVEs (OSV.dev), and license policy → allow / warn / block with a plain-English reason. The agent calls this before it writes the dependency.
Audito is a Model Context Protocol server. Your coding agent calls it as a tool — checking packages, auditing lockfiles, and fixing CVEs without you ever leaving the chat.
No SDK to wire up and no CI step required — just tools your agent already knows how to call.
check_packageValidate one package before adding it.
Typosquat, deprecation, known CVEs (OSV.dev), and license policy → allow / warn / block with a plain-English reason. The agent calls this before it writes the dependency.
audit_lockfileSweep an entire lockfile in one call.
Hand Audito a lockfile in any of 8 formats (npm, pnpm, yarn classic + berry, requirements.txt, poetry, Pipfile, Cargo, go.sum) and get back every finding at once.
suggest_actionsGet the exact fix per CVE.
The nearest safe upgrade: minimal bump, operator-preserving manifest edit, the per-ecosystem install command, and a breaking-change flag — everything the agent needs to apply the fix.
Add Audito as a remote MCP server and authenticate with an audt_… API token. Cursor and Claude Desktop connect the same way over mcp-remote.
# add Audito to Claude Code
claude mcp add audito \
--remote https://mcp.audito.dev \
--header "Authorization: Bearer audt_…"The same MCP endpoint works from Cursor and Claude Desktop via mcp-remote. Tokens scope to your workspace and can be revoked anytime.
Audito turns the agent's blind install into a reviewed one — and hands it the fix when something's already wrong.
The agent reaches for a package; Audito blocks the typosquat and names a safe alternative.
Likely typosquat of colors. Registered 6 days ago, 0 weekly downloads, no repo. Try chalk instead.
For what's already installed, Audito returns the minimal, operator-preserving bump per finding.
Create a token, add the server, and the next package your agent installs gets vetted first.