GitLab MCP includes large, unrequested payloads

[jjarokergc]

GitLab MCP produces verbose response payloads that increase avoidable, main-loop costs. The tool is returning everything:

• get_issue/update_issue response embedded the full milestone description
• get_merge_request embedded a deployment_summary of 10 deployment records with nested pipeline objects
• update_issue additionally echoes the entire updated description back

In the past session, I estimate 120–180k tokens of repeated boilerplate.

I host gitlab-mcp remotely. My configuration is:

    environment:
      SSE: true
      STREAMABLE_HTTP: false
      REMOTE_AUTHORIZATION: false
      HOST: 0.0.0.0
      PORT: 3002
      GITLAB_API_URL: "https://gitlab.example.com/api/v4"

      GITLAB_TOOLSETS: all

Would you please tell me if my gitlab-mcp is misconfigured.

[jjarokergc]

I benchmarked gitlab-mcp against glab. Claude ai using glab versus gitlab-mcp produced 98% reduction in token usage.

Before/after token cost — the three named offenders

Approximate tokens = characters / 4 (rough JSON proxy). "gitlab-mcp (full)" is the full-object payload the MCP tools return; "glab" is the field-selected read.

Offender	gitlab-mcp (full)	glab (field-selected)	Reduction
Milestone description echo — get_issue embeds the whole milestone object every call	6,237 chars / ~1,559 tok	146 chars / ~36 tok	~98%
deployment_summary — get_merge_request embeds 10 nested deployment records	~16,833 chars / ~4,208 tok¹	175 chars / ~43 tok	~99%
Echoed body — get_issue/update_issue returns the entire body to read/tick one checkbox	9,925 chars / ~2,481 tok	167 chars / ~41 tok	~98%

¹ MR full = 7,364 chars REST object + 9,469 chars deployment_summary embed (the trimmed 10-record shape gitlab-mcp actually returns, measured directly). glab's REST MR object carries no deployment_summary at all.

[zereight]

@jjarokergc how about now?