feat(genx): GenX connector — capacity analysis, diurnal plots, SLURM submission (integrates #51)

[qian-harvard]

Integrates #51 by @lauraahwa, bringing the new GenX connector up to the standard the rest of the distribution holds to. The author's commits are preserved as a parent of the merge.

Stacked on #58 (ANDES) — merge that first; the base will retarget to main automatically.

#51 merged cleanly but was inert: it imported the mcp 1.x API, appeared in no packaging or registry list, and could not start without a .env file.

Made it able to start

• server.py imported mcp.server.fastmcp, removed in mcp 2.0. This repo pins mcp>=2,<3, so the server raised ModuleNotFoundError before registering a single tool.
• slurm.py resolved GENX_DIR at import and raised RuntimeError when unset, so import GenX.server failed on any machine without a .env — taking down the plotting tools, which need no cluster at all. Everything resolves at call time now.
• Dropped the .env / python-dotenv mechanism for the repo's own ~/.powermcp/config.toml, read through powermcp.config.get_path, with GENX_DIR still honoured as an override. That removes a parallel config mechanism and a dependency the base install does not carry.

Made it ship

• Registered in powermcp/registry.py with a genx.repo_root config key, added to the wheel force-include and the sdist include list, and given a genx extra. Without these the directory was in neither the wheel nor the sdist, and powermcp run genx did not exist.
• Added to tests/test_sdk_imports.py SERVER_DIRS — the sweep that would have caught the mcp 1.x import in the first place.

Security and correctness

• Shell injection. case_name was interpolated unescaped into the bash script piped to sbatch, in both an #SBATCH directive and inside echo "Case: ...". A name closing the quote and opening a command substitution executed on the cluster under the user's own account. Job names are now restricted to [A-Za-z0-9._-] and every interpolated value is shlex.quoted.
• No path containment. No argument went through powermcp.sandbox, the policy every other connector follows. csv_path and the plot outputs are checked directly; case_dir and scenario_path are checked after resolution, since either may be relative to the GenX directory and only the resolver knows the path that actually gets opened.
• stdout pollution. print() in diurnal_generation and plot_capacity wrote into the stdio JSON-RPC stream. Routed to logging.
• summarize_capacity returned a pandas DataFrame from a -> dict tool.
• load_capacity_csv printed on a missing required column and then walked into the KeyError two lines later.
• Three of six tools had no exception handling, so ordinary input errors escaped as raw MCP protocol errors while the others returned {"success": False}. One shape now.
• compute_capacity_cost silently skipped an unknown CapRes region and reported a successful $0 capacity price, and divided by a zero peak demand.
• sbatch gained a timeout — a wedged SLURM controller froze the whole server.

Verification

• tests/test_genx_server.py, 25 new tests: script generation and injection refusal, configuration resolution, capacity CSV analysis, the shared error shape, and containment.
• Full suite: 408 passed, 17 skipped.
• python -m build → GenX/ confirmed present in both the wheel (powermcp/_servers/GenX/) and the sdist; twine check passes.
• powermcp list shows genx; powermcp doctor reports ok with set: genx.repo_root.

Not addressed

plot_capacity.classify_resource and diurnal_generation.classify are two independent resource classifiers that disagree on petroleum and distributed generation. Real, but unifying them changes plot output and belongs with the author — filed as a follow-up.

Closes #51.

🤖 Generated with Claude Code