Prepare repo for public release

[will-fawcett]

Summary

Brings this repo up to the same public-release standard as eve-mcp — packaging, CI, formatting, badges, tests — and fixes a batch of bugs found while reading every file.

Bugs fixed

• auth.py — function-arg key_path was being clobbered to None (so the documented "tried first" behaviour never happened); env-var fallback was os.getenv("GEE_SERVICE_ACCOUNT") or os.getenv("GEE_SERVICE_ACCOUNT") (duplicated; now also tries EE_SERVICE_ACCOUNT); the early GEE_PROJECT raise short-circuited levels 3/4 of the auth chain; the hardcoded geo-stars-…json filename has been replaced with a GEE_KEY_PATH env var; added GEE_SKIP_AUTH=1 for tests.
• codegen.py — typos thiking_python, ptyhon_code_thinking.
• analysis.py — stray ~ at EOF; _extract_factuality_issues was defined twice.
• coderun.py — try: del(gee_main) was a no-op (operates on local scope).
• helpers.py — extract_tag regex was greedy, which crossed subsequent ``` fences; extract_xml_tag did not reject a missing opening tag.
• tools_analysis.py — _assess_factuality_issue was registered with a leading underscore and a 2-arg signature (the inner helper takes 6 args), so the tool was effectively broken. Now registered as assess_factuality_issue with the correct signature.
• utils.py — duplicated several helpers and the SensitivityAnalizer class from analysis.py. Trimmed to the two metadata fns actually used.
• genai.py — default model gemini-3.1-pro-preview is unstable; bumped to gemini-2.5-pro.
• client.py — test() declared -> str but returned a dict.
• example.ipynb — committed personal project IDs (ee-raulramos, genai-dev-454121) and embedded tile URLs from a previous run; cleared and replaced with placeholders.
• server.log — runtime log was checked in.

Restructure

• Moved server/ → src/gee_mcp/server/ (eve-mcp layout).
• Renamed FastMCP server mcp-gee-server → gee-mcp.
• __main__.py exposes a main() function (testable).
• Removed file-based logging from app.py (was writing to a path that no longer made sense after the move and is now gitignored anyway).

Packaging & tooling

• pyproject.toml (Poetry) replacing requirements.txt, supporting Python 3.11–3.14.
• MIT LICENSE.
• .gitignore, .pre-commit-config.yaml, .pylintrc, .secrets.baseline ported from eve-mcp. Coverage gate is --fail-under=50 (eve-mcp's 100 isn't realistic on this surface area without much heavier mocking).
• .github/workflows/main.yml running pre-commit on the Python matrix.

Tests

• tests/ with 23 tests, all passing locally:
  • test_app.py — FastMCP instance + name.
  • test_main.py — main() entry point invokes mcp.run(transport="stdio").
  • test_init.py — all 23 tools registered with the FastMCP app.
  • test_helpers.py — extract_xml_tag, extract_tag (incl. non-greedy regression), remove_leading_spaces.
  • test_models.py — Pydantic cross-field validation for RegionParams, ComputeIndexParams, MultiPeriodParams.

README

• CI / Python / License badges (pointing at FrontierDevelopmentLab/gee-mcp).
• Alpha status callout.
• Updated install (Poetry), run (python -m gee_mcp.server), and config sections; new GEE_KEY_PATH / GEE_SKIP_AUTH env vars documented.
• Typos fixed (anem, Notebookes, issies, generate_generate_…).
• assess_factuality_issue tool now documented (it was missing).

Test plan

• python -m pytest tests — 23 passed locally.
• All 23 tools register at FastMCP startup (verified via FastMCP list_tools).
• Every .py file AST-parses cleanly.
• No personal project IDs remain in the notebook.
• grep -rn "geemcp\|rramosp" returns only the (legitimate) rramosp author entry in pyproject.toml.
• CI green on push (Python 3.11–3.14).
• Run poetry run pre-commit run --all-files after first poetry install to generate poetry.lock.

Notes for follow-up (out of scope)

• poetry.lock is not committed yet — generated on first poetry install.
• Lazy GEE auth would be cleaner than the current import-time setup_gee() + GEE_SKIP_AUTH=1 test override; full conversion to lazy init touches every tool and was deferred.
• Coverage starts at 50% — the GEE/Gemini/HTTP surface is large; raising it requires heavier mocking.