Docker MCP Catalog and Toolkit

Discover, run, and manage MCP servers as Docker containers.

Freemium· Docker Personal: $0 · Docker Pro: $11 · Docker Team: $16 · Docker Business: $24MCP Servers

Best for

Developers and teams already on Docker Desktop who want a secure, catalog-driven way to plug MCP servers into Claude, Cursor, VS Code, and other AI clients without hand-managing processes and secrets.

Skip if

Users on lightweight setups who want a single MCP server with no container runtime, or air-gapped environments that cannot pull from Docker Hub.

Docker MCP Catalog and Toolkit is Docker's official platform for discovering, running, and distributing Model Context Protocol (MCP) servers as containers. It has three main parts: the MCP Catalog on Docker Hub (a curated registry of hundreds of pre-built MCP servers from publishers like GitHub, Stripe, MongoDB, Elastic, Grafana, New Relic, and Neo4j); the MCP Gateway (an open-source Docker CLI plugin that runs, isolates, and multiplexes MCP servers behind a single endpoint); and the MCP Toolkit inside Docker Desktop, which wires those servers into local AI clients like Claude Desktop, Cursor, VS Code, Windsurf, Continue, and Goose with a one-click setup. Because every server runs in its own container with minimal host privileges, developers get sandboxed execution instead of dumping npx/uv processes onto their laptop, and secrets are centralised through Docker Desktop's credential store rather than pasted into per-client JSON configs. The Gateway aggregates many servers behind one MCP endpoint, so an agent sees a consolidated tool list and can be filtered, rate-limited, and traced from one place. Typical workflows include letting Claude query a Postgres database, giving Cursor safe shell/file access, wiring an agent into GitHub, Jira, or Stripe, or running a small internal catalogue of company-specific MCP servers for a team. Publishers can also submit their own images to the catalog to reach the growing MCP ecosystem. For anyone already running Docker Desktop, it is the lowest-friction way to adopt MCP without hand-managing a dozen server processes.

Editor's take

This is the most credible attempt yet to make MCP feel like installing an app instead of babysitting a Python process. If you already live in Docker Desktop, the Toolkit plus Gateway removes almost all of the setup pain, and containerised isolation is the right default for tools that can touch your shell, files, and API keys. The catalog still needs editorial pruning, but the direction is clearly right.

— The AI Tool Bible editorial team

Pros

✅ Largest curated catalog of containerised MCP servers, with images from major vendors (GitHub, Stripe, MongoDB, Elastic, Grafana, Neo4j and more).
✅ Every server runs sandboxed in a container with least-privilege defaults, avoiding the security foot-guns of running raw npx/uvx MCP processes.
✅ One-click enable/disable in Docker Desktop's MCP Toolkit UI, no hand-editing JSON config for each AI client.
✅ MCP Gateway multiplexes many servers behind a single endpoint with built-in logging, call tracing, and OAuth helpers.
✅ Central Docker Desktop secret store means credentials are entered once and reused across Claude, Cursor, VS Code, Windsurf, and others.
✅ Gateway and registry are open source under MIT, so teams can self-host or customise the plumbing.
✅ Enterprise controls through Docker Business (image access policies, registry access management) make it viable for regulated shops.

Cons

⚠️ Requires Docker Desktop (or at minimum a Docker Engine) — extra weight if you only wanted a single Python MCP server.
⚠️ Container cold-starts add noticeable latency versus running an MCP server directly on the host.
⚠️ Catalog quality is uneven: some third-party servers are thin wrappers or lag behind their upstream projects.
⚠️ Advanced enterprise features (private catalogs, RBAC, image access management) sit behind Docker Business pricing.
⚠️ Windows and Linux workflows are less polished than macOS; some Toolkit features assume Docker Desktop UI, not headless engines.
⚠️ Still a moving target — MCP itself is young and the Toolkit's UX, config format, and CLI flags change between Docker Desktop releases.

Use cases

Wiring Claude Desktop into Postgres, GitHub, and StripeGiving Cursor sandboxed shell and filesystem accessRunning a private team catalog of internal MCP serversAggregating many MCP servers behind one Gateway endpointCentralised secret management for AI tool credentialsOAuth-authenticated MCP servers for SaaS APIsPublishing a company's own MCP server to Docker HubLocal agent development with logging and call tracingEnterprise governance over which MCP tools developers can enable