Skip to content

Geo-Insights MCP

The work of adapting a city to heat and rain is done in spreadsheets that nobody keeps current. A municipality needs to know what a sedum roof actually retains, what a permeable pavement costs per square metre, which of forty-odd measures its own policy documents require, and what the temperature was at the nearest station last Tuesday. That information exists, but it lives in a national weather API, three sensor networks with different units, a water board's PDF and a procurement file.

Geo-Insights is a Dutch service that puts an environmental data layer and a blue-green infrastructure catalog behind one remote MCP server, keyless, so an agent can query all of it in one place. It is the first entry in this catalog for urban climate adaptation, a term that returns zero hits across every guide, heading and sweep note here.

The honest finding of this pass is a split: the catalog half answers and the policy half does not, and the sensor half was unreachable for the whole session.

Server type:   remote Streamable HTTP, JSON-RPC 2.0, session header issued
Endpoint:      https://mcp.geo-insights.nl/mcp
Server:        "Geo-Insights Environmental Data", protocol 2025-06-18,
               capabilities: tools, resources, prompts, version field empty
Registry:      io.github.mathisvfr/environmental-data 1.2.0, streamable-http,
               same URL, status active, published 2026-11-14
Auth:          none, and the instructions string says so
Tools:         12
Scope:         Dutch sensor networks plus a 44-measure blue-green catalog

The evidence tier

Keyless and drivable, which is the best tier available: initialize answered 200 with a session id and no credential, and every tool below was called at least once from a fresh session.

Five tools answered with data. Seven refused, and the refusals name their own cause:

Tool Result on 2026-11-14 Cause
list_sources answered 4 sources returned
get_parameters answered 36 parameters
search_bgi_measures answered measure list returned
get_measure_detail answered full measure record
compare_measures answered 2 to 5 measures side by side
get_latest_readings refused sensor_data_unavailable, http_400
get_data_freshness refused sensor_data_unavailable, http_400
get_readings_history refused arguments require start and parameter
lookup_municipality refused catalog_unavailable, http_406
get_policy_requirements refused catalog_unavailable, http_406
rank_measures refused catalog_unavailable, http_406
get_context_summary refused catalog_unavailable, http_406

What the catalog half returns

search_bgi_measures returns measures with a 0 to 100 score per goal - flood, heat, drought, biodiversity, social, economic - each carrying a confidence field of computed, plus data_completeness and an updated_at. A sedum roof scores 75 on drought and 65 on heat against 30 on biodiversity and 12 on social; street trees score 80 on biodiversity and 70 on economic but only 18 on drought. That is the shape of a tool that is telling a municipality which goal each purchase serves rather than which measure is best.

get_measure_detail adds the numbers an engineer asks for. For the lightweight sedum roof:

Field Value
Retention capacity 101.6 mm typical, 76.2 to 152.4 mm range
Peak flow reduction 75%
Storage volume 0.064 m3/m2 (minimum)
Typical lifespan 15 years
Data completeness 88.8
Reference family Building Solutions

compare_measures puts the cost side next to it: capex_typical 50.0 and opex_typical 3.0 for the sedum roof against an extensive green roof at a lower capex, with retention, cooling effect in degrees C and infiltration rate in mm/hr on the same row.

One caution on those figures: capex_typical arrives without a currency or a unit. The number is useful for ranking measures against each other and should not be quoted as a price per square metre until the vendor states the unit.

Why the policy tools matter more than the sensors

The reason to reach for this server is not the temperature at a station, which a national weather API gives you anyway. It is get_policy_requirements, which claims to extract a municipality's blue-green obligations from its own official publications, and rank_measures, which is supposed to weight the 44-measure catalog by a municipality's policy profile.

Both of them, and the municipality lookup they depend on, answered catalog_unavailable with http_406 on every attempt across two separate sessions, spaced minutes apart, while search_bgi_measures and get_measure_detail answered normally before, between and after them. One catalog tool working while another reports the catalog unreachable is a partial backend, not a credential problem: there is no key to add and no scope to grant.

rank_measures also cannot be reached by argument alone: it needs a numeric gemeente_code, the only source of which is lookup_municipality, the tool that is down. So the recommendation layer is unreachable end to end today, even though the measure table underneath it is intact and readable.

{"error": "catalog_unavailable",
 "message": "BGI catalog data source is temporarily unreachable",
 "side_effects": "none",
 "cause": "http_406",
 "fix": "Retry after 30 seconds",
 "retry_after_seconds": 30}

The retry_after_seconds: 30 was not honoured by the service on retry. Recorded as a defect rather than as a temporary blip, because the same envelope came back on every call.

What the sensor half claims and what it returns

The server's own instructions describe ten sources and 35 normalized parameters. Measured:

  • list_sources returned 4 sources: Sensor.Community (807 stations, 1,862 datastreams, ODbL), Ohnics (44 stations), KNMI weather (46 stations, 460 datastreams) and Luchtmeetnet (103 stations, 0 datastreams).
  • get_parameters returned 36 parameters, not 35, and includes alias duplicates: temperature and air_temperature both read Air temperature °C, and humidity and relative_humidity both read Relative humidity %.

So a model that takes the instructions at face value will overstate the coverage of the sensor layer by a factor of two and a half, and a model that takes the parameter list at face value will treat two names for one measurement as two measurements.

Argument names that differ from the descriptions

Three tools reject the argument the description suggests, and each rejection is a pydantic validation error rather than a helpful message:

Tool Rejected Accepted
get_measure_detail measure_slug measure_id_or_slug
get_policy_requirements municipality gemeente_code_or_name
rank_measures gemeente_code_or_name gemeente_code

An agent that reads the tool description and sends the obvious name gets a validation error with no hint about which name is right; only tools/list carries the schema. Worth knowing before concluding a tool is broken.

What an operator should use it for

  • A green-roof or permeable-paving contractor quoting in the Netherlands: the measure catalog gives defensible per-goal scores and engineering ranges to put in a proposal, and the numbers come from the tool rather than from a marketing page.
  • A sustainability officer building a shortlist before a council paper: search_bgi_measures and compare_measures are enough to rank options on the six goals, as long as the ranking is done by the human and not by rank_measures, which is down.
  • Anyone needing actual weather or air-quality readings: not this server today. The sensor tools returned sensor_data_unavailable throughout.

What does not work

  • The policy and recommendation chain - lookup_municipality, get_policy_requirements, rank_measures, get_context_summary all refuse with http_406.
  • All sensor reads - get_latest_readings and get_data_freshness refuse with http_400.
  • get_readings_history requires start and parameter, neither of which appears in the description.
  • Version drift - the server reports an empty version string while the registry record is 1.2.0.
  • Coverage overstatement - 10 claimed sources, 4 returned; 35 claimed parameters, 36 returned with alias pairs.
  • Unitless cost - capex_typical and opex_typical carry no currency.

Frequently asked questions

Does Geo-Insights MCP need an API key?

No. initialize and every tools/call below answered without a credential, and the server's instructions state that no authentication is required.

Is this the same as a national weather API?

No. KNMI and Luchtmeetnet are two of its upstream sources; the added value is the blue-green measure catalog with costs, performance ranges and per-goal scores, which no weather API carries.

Can an agent rely on the scores being current?

Partly. Each parameter in get_measure_detail carries its own confidence, and each measure carries updated_at; the dates read 2025-12-28 and 2026-02-27 across the measures sampled here, so the catalog is a curated reference rather than a live feed.

What would make this a full entry rather than a partial one?

The four http_406 tools answering, and the sensor tools returning a reading instead of sensor_data_unavailable. The schema, the endpoint and the measure data are already sound.

Limits recorded

Dutch jurisdiction throughout, in Dutch or English field names. The measure catalog is a reference table, not a live price feed, and the two halves of the product fail independently. No credential is needed, so nothing here is gated - the only blocker is the vendor's own backend.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing