Docker MCP Toolkit
Author / channel: WorldofAI Format: video Source: Original Published: 2025-06-14 Sponsor: Docker (disclosed)
Summary
Docker shipped an MCP Toolkit extension for Docker Desktop that solves the three big pain points of the MCP ecosystem: discovery (curated catalog of 100+ verified servers), security (containerized isolation, secrets vault, no plaintext credentials in client config), and install effort (one-click enable + auto-wire to clients like Claude Desktop, Cursor, Continue, Gordon). Demos the GitHub MCP installed in 2 clicks and used from Cursor to create a repo. Recommends Context7 (live docs) and Desktop Commander (file/terminal ops) as must-have MCPs.
Key Points
- The ecosystem problem — pre-toolkit MCP install was: clone repo, install deps, configure secrets in plaintext JSON, no audit logs, no policy enforcement, no provenance. Toolkit solves all of this.
- Containerized isolation — every MCP server runs in its own container with restricted host access. Credentials stored in Docker’s secrets vault, not in the client config file.
- Curated catalog — 100+ verified MCP servers in the Docker MCP Catalog. GitHub, MongoDB, PostgreSQL, Slack, Notion, etc. One-click enable.
- Client integration — toolkit auto-wires enabled MCPs into Claude Desktop, Cursor, Continue Dev, Gordon, VS Code with Cline. The same enabled MCP is exposed to all configured clients simultaneously.
- Recommended MCPs from the demo:
- Context7 — live up-to-date documentation injection (works around model knowledge cutoff)
- Desktop Commander — file search/edit + terminal commands with codebase-wide context awareness
- GitHub MCP — PRs, tags, issues — natural-language git ops
Sponsorship & Bias Notes
Sponsored by Docker — disclosed openly. The toolkit is genuinely useful and the security framing is accurate (the prior model of pasting tokens into JSON config was a real risk). Take the “100x productivity” framing with the usual skepticism; the categorical improvement is real, the multiplier is marketing.
Connected Pages
- mcp — toolkit is the canonical install/distribution layer for MCP
- docker-model-runner — sibling Docker AI initiative
- claude-code, cursor, continue, cline — supported clients
- WorldofAI — channel hub