JFrog MCP Server

Official MCP server for the JFrog Platform: manage repositories, builds, runtime clusters, projects, package vulnerabilities, and artifact security scans.

Developer Tools by JFrog Bearer Token active

GitHub Docs npm: @jfrog/mcp-jfrog

Overview

The JFrog MCP Server is an experimental Model Context Protocol implementation maintained by JFrog that connects AI assistants to the JFrog Platform API. It exposes 22 tools covering repository management (local, remote, virtual), Artifactory Query Language (AQL) execution, build tracking, runtime cluster and container image monitoring, project and environment administration, public package intelligence, and security scanning of artifacts.

The server can run via stdio for local clients like Claude Desktop or Cursor, or in Server-Sent Events (SSE) mode for remote/multi-client setups. SSE mode adds connection ID tracking, automatic reconnection with exponential backoff, a /health endpoint, and structured logging. It is distributed as an npm package, a Docker image, and via Smithery.

Note: JFrog also offers an officially hosted, remote MCP server for production use. This open-source repo is marked experimental, but is the canonical self-hostable implementation maintained under the jfrog GitHub org.

Tools

Tool	Description
`check_jfrog_availability`	Check whether the configured JFrog platform instance is reachable and responsive.
`create_local_repository`	Create a local Artifactory repository.
`create_remote_repository`	Create a remote repository proxying an upstream source.
`create_virtual_repository`	Create a virtual repository aggregating multiple repos.
`list_repositories`	List repositories on the platform with optional filters.
`set_folder_property`	Set properties on a folder, optionally applied recursively.
`execute_aql_query`	Run an Artifactory Query Language (AQL) query.
`list_jfrog_builds`	List all builds tracked in Artifactory.
`get_specific_build`	Retrieve details for a specific build.
`list_jfrog_runtime_clusters`	List runtime clusters monitored by JFrog Runtime.
`get_jfrog_runtime_specific_cluster`	Get details for a specific runtime cluster.
`list_jfrog_running_images`	List running container images across runtime clusters.
`list_jfrog_environments`	List configured environments on the platform.
`list_jfrog_projects`	List projects on the platform.
`get_specific_project`	Get detailed information about a specific project.
`create_project`	Create a new project with administrative settings.
`jfrog_get_package_info`	Retrieve public package metadata across PyPI, npm, Maven, Go, NuGet, Hugging Face, RubyGems.
`jfrog_get_package_versions`	Get the version history and publication dates for a package.
`jfrog_get_package_version_vulnerabilities`	List known vulnerabilities for a specific package version.
`jfrog_get_vulnerability_info`	Look up detailed information for a CVE or vulnerability ID.
`jfrog_get_package_curation_status`	Check whether a package is approved, blocked, or inconclusive under curation policies.
`jfrog_get_artifacts_summary`	Generate a security issues summary for artifacts by severity (Low, Medium, High, Critical, Unknown).

Setup Guide

Prerequisites

Node.js v18 or higher (or Docker)
A valid JFrog platform instance (e.g. https://your-instance.jfrog.io)
A JFrog access token or identity token with sufficient permissions

Install via Smithery (recommended for Claude Desktop)

npx -y @smithery/cli install @jfrog/mcp-jfrog --client claude

Local build

git clone git@github.com:jfrog/mcp-jfrog.git
cd mcp-jfrog
npm i && npm run build

Docker

docker build -t mcp/jfrog -f Dockerfile .

Claude Desktop config (npm via github)

{
  "mcpServers": {
    "MCP-JFrog": {
      "command": "npm",
      "args": ["exec", "-y", "github:jfrog/mcp-jfrog"],
      "env": {
        "JFROG_ACCESS_TOKEN": "YOUR_TOKEN",
        "JFROG_URL": "https://your-instance.jfrog.io"
      }
    }
  }
}

Claude Desktop config (Docker)

{
  "mcpServers": {
    "jfrog": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "-e", "JFROG_ACCESS_TOKEN", "-e", "JFROG_URL", "mcp/jfrog"],
      "env": {
        "JFROG_ACCESS_TOKEN": "YOUR_TOKEN",
        "JFROG_URL": "https://your-instance.jfrog.io"
      }
    }
  }
}

Environment variables

Variable	Required	Purpose
`JFROG_ACCESS_TOKEN`	Yes	Authentication credential
`JFROG_URL`	Yes	Platform instance URL
`TRANSPORT`	No	`stdio` (default) or `sse`
`PORT`	No	SSE server port, default 8080
`CORS_ORIGIN`	No	CORS allowlist for SSE, default `*`
`LOG_LEVEL`	No	DEBUG, INFO, WARN, ERROR (default INFO)
`MAX_RECONNECT_ATTEMPTS`	No	SSE reconnection attempts, default 5
`RECONNECT_DELAY_MS`	No	Base delay between SSE reconnects, default 2000

Use Cases

Provision new local, remote, or virtual Artifactory repositories from a chat prompt during onboarding.
Investigate build provenance by listing builds and pulling details for a specific build name and number.
Audit container security across runtime clusters by listing running images and generating artifact issue summaries by severity.
Check whether a third-party npm, PyPI, or Maven package is approved by curation policy and surface its known CVEs before adoption.
Run AQL queries to find artifacts by properties (e.g. outdated SHA, missing license tag) and bulk-set folder properties.

Example Prompts

"Create a virtual npm repository called npm-virtual that aggregates npm-local and npm-remote."
"List all builds for the payments-service and show details for build number 482."
"Show me running container images across our prod runtime cluster and flag any with Critical vulnerabilities."
"Is lodash@4.17.20 curated as approved in our platform, and what CVEs affect it?"
"Run an AQL query for all artifacts in libs-release-local modified in the last 7 days."

Pros

Maintained under the official jfrog GitHub org with broad coverage of 22 tools across repos, builds, runtime, projects, packages, and security.
Supports both stdio and SSE transports, plus npm, Docker, and Smithery distribution.
Covers package and CVE intelligence across multiple ecosystems (PyPI, npm, Maven, Go, NuGet, Hugging Face, RubyGems), useful for supply-chain workflows.

Limitations

Marked experimental by JFrog, with a separate officially hosted remote MCP server recommended for production.
Requires a JFrog Platform instance and access token; not usable without a paid/configured tenant.
Tool input schemas are documented sparsely in the README, so some calls require trial-and-error or reading source.

Alternatives

Sonatype Nexus - alternative artifact repository; no official MCP server, community implementations only.
GitHub MCP server - for source and package management on GitHub Packages.
JFrog's officially hosted remote MCP server, referenced from JFrog Platform docs, for production use without self-hosting.