v0.3.0

a large rewrite of the MCP server with new tooling, testing, and observability.

Highlights

• MCP tool annotations — all 19 tools now declare read/write intent (READONLY, WRITE_IDEMPOTENT, WRITE_CREATE, etc.)
• MCP resources & prompts — 3 resources (categories, accounts, institutions) and 4 guided prompts (spending analysis, budget review, financial health check, transaction categorization)
• Log analyzer & session evaluator — scripts/analyze_logs.py and scripts/eval_session.py for measuring token usage and identifying optimization opportunities
• Token-optimized responses — compact categories (just id+name), lean bulk update responses, smarter compact transaction format
• Local CI script — uv run python scripts/ci.py runs the same checks as GitHub Actions

Server changes

• Consolidated logging — removed redundant analytics tracking, simplified to structlog
• Deleted ~200 lines of dead code (unused Pydantic models, duplicate helpers)
• Bulk transaction updates return compact {transaction_id, status} instead of echoing full objects
• get_transaction_categories returns compact {id, name} pairs by default (verbose=True for full)
• Compact transactions now include categoryId, only include pending/notes when present

Testing & CI

• 151 tests across 13 test files (up from ~70)
• GitHub Actions CI on Python 3.10–3.13 (ruff, mypy, pytest)
• Ruff linting and formatting enforced
• Code review workflow with sticky comments

Dependencies

• mcp 1.18 → 1.26
• monarchmoneycommunity 1.2 → 1.3
• pytest 8 → 9
• starlette 0.48 → 0.52, typer 0.19 → 0.24, uvicorn 0.38 → 0.41
• All deps bumped to latest

Docs

• README rewritten — cleaner structure, proper fork attribution to @colvint and @hammem
• PII policy added to CLAUDE.md — no real financial data in code or docs
• All examples use anonymized data