Skip to content

ci: stop dependabot re-proposing the uninstallable mcp 2.x bump - #24

Merged
Robobc merged 1 commit into
mainfrom
ci/dependabot-hold-mcp-major
Aug 19, 2026
Merged

ci: stop dependabot re-proposing the uninstallable mcp 2.x bump#24
Robobc merged 1 commit into
mainfrom
ci/dependabot-hold-mcp-major

Conversation

@Robobc

@Robobc Robobc commented Aug 19, 2026

Copy link
Copy Markdown
Contributor

What

Every weekly python-deps batch carries mcp==2.0.0, and every one is uninstallablestrands-agents, claude-agent-sdk and langchain-mcp-adapters all still declare mcp<2.0.0:

$ pip install --dry-run strands-agents==1.52.0 mcp==2.0.0
ERROR: ResolutionImpossible

Because the batch is grouped, that one bad pin takes 24 good bumps down with it — which is exactly how #19/#20 became unmergeable and had to be split by hand into #21#23.

Changes

  • Hold mcp at 1.x via ignore on major updates only (minor and patch still flow). The comment explains the reasoning and names the one-line check for deciding when to lift it, so it reads as a deliberate hold rather than a forgotten pin.
  • Fix the stale file header — it claimed GitLab is the source of truth and that bumps arrive here via export, which stopped being true when this repo became primary. Kept the part that still matters: CI here runs no container builds or agent invocations, so a green check on a dependency PR means little, and grouped batches should be split by risk.

Verified

YAML parses, 3 update blocks intact, ignore rule reads back as expected.

Follow-up

#20 can be closed — it was generated against the pre-#21/#22/#23 base, so it re-reverts everything just merged and still carries the bad mcp pin. With this in place, the next scheduled batch should be small and mergeable.

Every weekly python-deps batch has carried mcp==2.0.0, and every one is
uninstallable: strands-agents, claude-agent-sdk and langchain-mcp-adapters
all still declare mcp<2.0.0 (pip: ResolutionImpossible). Worse, the single
bad pin drags the whole grouped batch down with it — that is how a 25-package
PR became unmergeable and had to be split by hand into #21-#23.

Holds mcp at 1.x via an ignore rule for major updates only (minor/patch still
flow). The comment says why, and names the one-line check for deciding when
to remove it, so this reads as a deliberate hold rather than neglect.

Also corrects the file header: it claimed GitLab is the source of truth and
that bumps arrive here by export, which stopped being true when this repo
became primary. The point that survives is the one that matters — CI here
runs no container builds or agent invokes, so a green check on a dependency
PR means little, and grouped batches should be split by risk.

Verified: yaml parses, 3 update blocks, the ignore rule reads back as
expected.
@Robobc
Robobc merged commit 837e085 into main Aug 19, 2026
6 of 7 checks passed
@github-actions

Copy link
Copy Markdown

Commit: 5fdd165 | Updated: 2026-08-19 00:11:09 UTC

Security Scan Results

Scanner S C H M L I Time Action Result Thresh
bandit 0 0 0 0 0 0 912ms 0 PASSED MED (g)
cdk-nag 0 0 0 0 0 0 7.1s 0 PASSED MED (g)
cfn-nag 0 0 0 0 0 0 12ms 0 PASSED MED (g)
checkov 0 0 0 0 0 0 5.2s 0 PASSED MED (g)
detect-secrets 0 0 0 0 0 0 114ms 0 PASSED MED (g)
grype 0 0 0 0 0 0 1m 1s 0 PASSED MED (g)
npm-audit 0 0 0 0 0 0 313ms 0 PASSED MED (g)
opengrep 0 3 0 0 0 0 16.6s 3 FAILED MED (g)
semgrep 0 0 0 0 0 0 <1ms 0 MISSING MED (g)
syft 0 0 0 0 0 0 2.5s 0 PASSED MED (g)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant