Rust workspace that scans repositories for SAFE-MCP techniques. It ships a CLI (safe-mcp-scan) and an MCP server (safe-mcp-analyzer) that call the shared engine.

• Prereqs: Rust stable, provider key in config or env (OpenAI/Anthropic), and schemas/technique.schema.json path if running outside the repo root.
• CLI example:

  cargo run -p cli -- --provider openai --model-name gpt-4o-mini \
    T1001 --repo /path/to/repo \
    --schema $(pwd)/schemas/technique.schema.json \
    --json [--llm-review]

• MCP server: cargo run -p server --bin safe-mcp-analyzer (configure providers/filters in YAML/JSON).
• Batch scans: ./run_scans.sh runs all specs under techniques/, continues on failures, writes scan_outputs/*.json.

• Specs: active techniques in techniques/ (top set), additional specs in techniques_backup/.
• SAFE-MCP corpus (gitignored): safe-mcp/README.md, safe-mcp/techniques/<ID>/README.md, safe-mcp/techniques/prioritized-techniques.md, mitigations under safe-mcp/mitigations/.
• Schema: schemas/technique.schema.json (pass --schema when running outside repo root).

• Prompts include file path/extension/line range, README excerpt, rule hints; temperature pinned to 0.
• Path filters: include/exclude globs/exts and max_file_bytes (0 disables); docs/manifests allowed by default.
• Optional --llm-review post-filters findings; non-fatal on failure; reuses configured OpenAI model/key.
• Evidence is mandatory (file, lines, snippet) for every finding; info-only findings do not fail the scan.

• See AGENTS.md for contributor workflow, commands to run, and coding/testing conventions.
• License: Apache 2.0 (see LICENSE). Don't commit changes to safe-mcp/ corpus.