Skip to content

FlowMCP v4.2 — Grading as a Versioned Standard

← All posts

FlowMCP v4.2 — Grading as a Versioned Standard

A schema is only as useful as it is clear. FlowMCP has always graded schemas — but until now the rules for how a schema is scored and graded lived as an internal protocol. With this release, grading becomes a published, versioned standard in its own right: the Grading-Spec v2.0.

Why this matters: a network of gradings

The point of a written, versioned grading standard is not the grade itself. It is reproducibility across people. When the rules — the eleven areas, the five-status model, the thresholds, the veto list — are public and versioned, a second person can grade a schema and arrive at a comparable result. Other maintainers and external contributors can grade by the same standard.

That is the groundwork for a contributor network: not a single team handing down grades, but an open standard that many people apply, producing a shared, comparable body of gradings. Grading stops being an internal tool and becomes something others can join.

A standard the AI can improve against

There is a second effect, and it is the one we are most excited about. When the criteria are written down — exactly which areas count, what a good schema looks like, where the typical weaknesses are — the AI itself knows what to aim for. It can read a schema, see where it falls short of the standard, and propose concrete improvements. The grade is no longer a verdict at the end; it becomes a checklist the AI works toward.

That turns the project into a self-improving loop: schemas are graded against explicit, shared rules, the AI uses those same rules to sharpen them, and a human can step in and improve them by the very same criteria. Everyone — people and AI — is reading from the same page.

FlowMCP delegates — it does not absorb

The FlowMCP Schemas-Spec (now v4.2) remains the highest instance — it defines what a schema, a selection, and the primitives are. What is new in 4.2 is a clean delegation: FlowMCP hands the grading model to a separate, independently versioned standard and points to it. The two specs have separate version numbers but are connected — the Schemas-Spec owns the upstream scoring transport, the Grading-Spec owns the grading model on top of it.

What’s new

  • v2, honestly numbered. The earlier 1.0/1.1 line was a short-lived experiment. The current break is real, so it is released as 2.0.0 rather than carrying a misleading minor number.
  • Eleven grading areas with a five-status node model and a derived rollup.
  • Delegation is the substance of the 4.2 bump — FlowMCP delegates grading to its own versioned sub-standard.
  • Its own docs area — Grading is navigation point 5, with its own version badge (v2.0), separate from the Specification badge.
  • CLI grading is experimental — usable today via the grading command area, but its surface may still change.
  • Peripheral modules follow — the grading reference implementation moves to a code-only repository; the single source of truth for the standard is the spec.

Where to read it

The Grading standard lives at /grading/ — its own area in the docs, versioned independently of the Schemas-Spec. The reference implementation is the flowmcp-grading repository (code only); the standard itself is the spec.

This is a first step. The standard is written down, versioned, and published — so the next gradings, wherever they come from, can speak the same language.

Related Posts

hackathon

Making the Connection — How FlowMCP Became a Mobility Framework

Story from DB InfraGO's 'Anschluss erreichen' hackathon, where FlowMCP's main contributor reached 3rd place with a multi-agent mobility assistant.

release

FlowMCP v4.0 — Skills, Selections, Pipes

How deterministic structures carry LLM composition without the AI hallucinating parameters.

release

FlowMCP v4.1 — GTFS as the First Data Class with Its Own Add-on

How the gtfs-sqlite-toolkit turns GTFS feeds into auditable SQLite resources and makes FlowMCP the first real mobility engine.