[mcp] fix: advertise gpu id item limit

[Wangmerlyn]

Summary

• add shared MAX_GPU_IDS for the public gpu_ids list-size contract
• advertise that limit as MCP start_keep.inputSchema.properties.gpu_ids.maxItems
• cover the runtime cap and MCP schema in tests, and align docs/AGENTS wording

Verification

• PYTHONPATH=src pytest tests/mcp/test_server.py -k mcp_tools_list_exposes_keepgpu_actions -q
• PYTHONPATH=src pytest tests/utilities/test_session_config.py -q
• PYTHONPATH=src pytest tests/mcp/test_server.py tests/utilities/test_session_config.py -q
• mkdocs build --strict
• PYTHONPATH=src pytest tests -q
• pre-commit run --all-files --show-diff-on-failure

Local subagent review: no must-fix issues found.

README note: Zenodo DOI badge remains present at README.md:5.

Summary by CodeRabbit

• New Features

  • GPU selection limits are now consistently enforced and exposed across the API, CLI, and MCP tool schema.
  • gpu_ids lists are capped at 64 entries, and this limit is now clearly advertised to clients.

• Documentation

  • Clarified GPU selection rules, including invalid list cases and startup behavior when no devices are found.
  • Updated API, CLI, and MCP guidance to reflect the 64-item limit.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 01e13d92-fa50-4533-9af1-50b1ab476dfd

📥 Commits

Reviewing files that changed from the base of the PR and between 1adedf8 and e7b3c2b.

📒 Files selected for processing (9)

• AGENTS.md
• docs/guides/mcp.md
• docs/guides/python.md
• docs/reference/api.md
• docs/reference/cli.md
• src/keep_gpu/mcp/server.py
• src/keep_gpu/utilities/session_config.py
• tests/mcp/test_server.py
• tests/utilities/test_session_config.py

---

📝 Walkthrough

Walkthrough

A MAX_GPU_IDS constant (64) is introduced in session_config.py and used by validate_gpu_ids for length validation. The MCP server's start_keep tool schema now advertises this limit via maxItems. Documentation and tests are updated accordingly.

Changes

MAX_GPU_IDS Enforcement

Layer / File(s)	Summary
Shared constant and validation logic src/keep_gpu/utilities/session_config.py, tests/utilities/test_session_config.py	Introduces MAX_GPU_IDS = 64 constant, updates validate_gpu_ids to compare against it instead of a literal, and adds a test verifying accept/reject behavior at the boundary.
MCP server schema wiring src/keep_gpu/mcp/server.py, tests/mcp/test_server.py	Imports MAX_GPU_IDS and applies it as maxItems on the start_keep tool's gpu_ids input schema; tests assert the schema exposes the correct maxItems value.
Documentation updates AGENTS.md, docs/guides/mcp.md, docs/guides/python.md, docs/reference/api.md, docs/reference/cli.md	Documents the 64-item gpu_ids limit, the maxItems schema contract, and related validation/telemetry behavior across guides and reference docs.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• Wangmerlyn/KeepGPU#59: Both PRs touch the MCP server's start_keep handling in src/keep_gpu/mcp/server.py, with this PR adding MAX_GPU_IDS-driven gpu_ids schema limiting on top of the server implementation from #59.
• Wangmerlyn/KeepGPU#146: Both PRs change start_keep's gpu_ids validation in the same server module, this PR adding a schema-level maxItems cap alongside prior validation logic.

Poem

A rabbit hopped through code so neat,
Counting GPUs, sixty-four a treat,
No more, no less, the schema's clear,
maxItems now sits right here,
Docs and tests all in a row—
Sixty-four's the limit, don't you know!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: advertising the GPU ID item limit in the MCP schema.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/mcp-gpu-ids-max-items

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}