Blog
RSS Feed →Updates, tutorials, and release notes for FlowMCP.
FlowMCP v4.2 — Grading as a Versioned Standard
FlowMCP Spec v4.2 delegates schema grading to its own independently versioned standard — the Grading-Spec v2.0, published as its own docs area so anyone can grade by the same rules.
Read more →What an agent's data lookup really costs — in tokens, dollars, and CO2
A repeated agent data lookup has a real, measurable cost. What it is, what drives token burn in tool discovery, and how pre-built schemas cut exploration cost to near zero.
token-economicsThe Amnesiac Agent: What a Pre-Built Schema Saves in Tokens, Dollars, and CO₂
Every AI agent wakes up with no memory of your API and re-derives it from scratch — every session, for everyone. A scripted, honest accounting of what a pre-built FlowMCP schema saves instead, measured on 47 real Berlin open-data sources, with ranges where we can only estimate.
geoOpen Data Germany × FlowMCP — A Tourism Graph You Can Combine
An experiment in joining Germany's national tourism knowledge graph with other open-data sources via the geo-dzt-toolkit — keyed access, renderable trail routes, a DHID bridge to public transit, an honest coverage contract, and a per-object Creative Commons licence model you must actually read.
data-formatsOpenStreetMap Around a Point — The Overpass Live-Query Add-on
How the geo-overpass-toolkit turns the Overpass API into a safe, reusable geo source — curated multi-key categories, pre-built combined selections, and one union query per call so rate limits never bite.
data-formatsCSV and TSV from a URL — With a Mandatory Parse Config
Why a CSV file can't describe itself, and how the geo-csv-tsv-toolkit loads geo CSV/TSV from a URL into memory by forcing every parse decision to be explicit — no silent defaults.
releaseFlowMCP v4.3 — Two Tracks: Development and Grading-Monitoring
FlowMCP Spec v4.3 splits the schema lifecycle into two tracks and aligns grading to the breaking Grading-Spec 3.0.0, where a failed pretest now emits a blocked record instead of aborting — alongside a leaner CLI, a single grading path, and four geodata add-ons.
data-formatsGeoJSON as a URL-Loaded In-Memory Resource
How the geo-geojson-toolkit fetches a complete GeoJSON FeatureCollection by URL, validates it on load, and lets FlowMCP auto-inject spatial query tools served from memory — without a single line of parsing config.
data-formatsSQLite — One Store for Local, Large, and Private Data
Not every dataset belongs behind a live API. FlowMCP has three ways to anchor data locally — preload, shared lists, and SQLite resources — and a clear rule for which to reach for.
data-formatsOParl — One Schema for Germany's Council Information Systems
How a single vendor-agnostic FlowMCP schema opens read access to German municipal council data — meetings, papers, people — across more than a hundred council systems.
releaseFlowMCP v4.1 — GTFS as the First Data Class with Its Own Add-on
How the geo-gtfs-toolkit turns GTFS feeds into auditable SQLite resources and makes FlowMCP the first real mobility engine.
releaseFlowMCP v4.0 — Skills, Selections, Pipes
How deterministic structures carry LLM composition without the AI hallucinating parameters.
hackathonMaking 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.