GraphCheck 0.2.0

GraphCheck 0.2.0

Changed

• Leaner base install. pip install graphcheck no longer pulls the LLM provider SDKs or the MCP server stack. Add them with extras when you need them:

  • pip install graphcheck[generate] — AI-assisted check authoring (graphcheck generate)
  • pip install graphcheck[mcp] — the agent surface (graphcheck mcp serve)

  Each command prints the matching install hint if its extra is absent.

Fixed

• graphcheck init no longer scaffolds a drift check that errors on the first run. A fresh project's default checks are baseline-free, so the first graphcheck run returns a clean, understandable result.

Upgrade

pip install --upgrade graphcheck

Full detail: see CHANGELOG.md.

GraphCheck 0.1.0

GraphCheck 0.1.0 — first public release

GraphCheck is semantic observability for property graphs — pytest for
knowledge graphs. Point it at a Neo4j database, declare checks in YAML,
and get pass/fail verdicts with evidence.

Install

pipx install graphcheck      # or: pip install graphcheck

Requires Python 3.12+ and a Neo4j database.

Highlights

• Command-line workflow: init, debug, profile, run, report, generate, redact.
• Built-in check packs: core conformance/competency/drift checks, plus a PII scan pack.
• Read-only by design: connects with a read-only credential and rejects writes.
• Offline HTML reports and a machine-readable results.json with a stable 0/1/2/3 exit-code contract.
• MCP agent surface: three tools let an AI agent list checks, run a suite, and read results.
• LLM-assisted check authoring (graphcheck generate) with a human approval gate.
• Redacted exports (graphcheck run --redact) with fail-closed literal verification.

Docs

Quickstart, CI setup, check reference, and troubleshooting live in the repo.

Full change detail: see CHANGELOG.md.