<![CDATA[πŸš€ New project: GitLab MCP Server]]>πŸš€ New project: GitLab MCP Server

Plug any AI assistant (Claude, Copilot, Cursor…) into the entire GitLab API via Model Context Protocol.

⚑ 1006 tools β€” full REST v4 + GraphQL
🧠 32 meta-tools to cut LLM token use
πŸ”’ Read-only & safe (dry-run) modes
🐳 Single static binary β€” zero deps
πŸ–₯️ Win/Linux/macOS Β· amd64 & arm64
πŸ” Self-hosted GitLab + TLS

Written in Go from scratch.

πŸ’» https://github.com/jmrplens/gitlab-mcp-server
πŸ“– https://jmrplens.github.io/gitlab-mcp-server/


]]>https://board.circlewithadot.net/topic/2e8170e8-b285-47ad-8d02-e2ce847142b8/new-project-gitlab-mcp-serverRSS for NodeThu, 14 May 2026 23:26:21 GMTThu, 30 Apr 2026 14:56:35 GMT60<![CDATA[Reply to πŸš€ New project: GitLab MCP Server on Thu, 30 Apr 2026 18:29:46 GMT]]>@dr41d45 Fair point! That's exactly why meta-tools are the default mode πŸ™‚

The LLM doesn't see 1006 tools β€” it sees 32 domain dispatchers (gitlab_issue, gitlab_mr, gitlab_pipeline…) each routing to the actual op via an 'action' parameter.

Full API coverage, ~32 tool definitions of context cost. Individual mode is opt-in.

Check docs: https://jmrplens.github.io/gitlab-mcp-server/tools/meta-tools/

]]>https://board.circlewithadot.net/post/https://mstdn.jmrp.io/users/jmrplens/statuses/116495075643820148https://board.circlewithadot.net/post/https://mstdn.jmrp.io/users/jmrplens/statuses/116495075643820148Thu, 30 Apr 2026 18:29:46 GMT<![CDATA[Reply to πŸš€ New project: GitLab MCP Server on Thu, 30 Apr 2026 16:47:28 GMT]]>@jmrplens

1006 tools to poison AI prompt? πŸ™‚

]]>https://board.circlewithadot.net/post/https://phpc.social/users/dr41d45/statuses/116494673429385148https://board.circlewithadot.net/post/https://phpc.social/users/dr41d45/statuses/116494673429385148Thu, 30 Apr 2026 16:47:28 GMT