Mixed-quality MCP server with 30 tools covering French open data APIs. All tools have descriptions (strong point) and explicit input schemas with type definitions. However, most descriptions are terse (avg 85 chars vs 194-char baseline), parameter descriptions are minimal or missing context about formats/constraints, and no output schemas are documented. Tool naming follows verb_noun convention consistently. Error handling is absent from visible code. No input validation guidance. The server prioritizes coverage breadth over definition depth, a common pattern in data aggregator servers, but falls short of production quality on parameter documentation and output specification.
Tools (30)
calculate_isochroneread onlysource verified77/100
Calculer une isochrone ou isodistance (zones accessibles en X temps/distance)
calculate_routeread onlysource verified77/100
Calculer un itinéraire entre deux points avec l'API IGN Navigation (voiture, piéton, etc.)
clear_cachewritesource verified65/100
Supprimer tous les fichiers cache
export_cached_dataread onlysource verified72/100
Exporter les données cachées au format JSON ou GeoJSON
get_cached_dataread onlysource verified65/100
Récupérer les métadonnées d'un cache par son ID
get_commune_inforead onlysource verified68/100
Obtenir les informations d'une commune par son code INSEE ou nom
Add formal output schema documentation for every tool. Specify field names, types (string, number, object, array), and required fields. Example for search_datasets: {type: 'object', properties: {results: {type: 'array', items: {type: 'object', properties: {id, name, description, organization, tags, ...}}}, total_count: {type: 'integer'}, ...}}
Upgrade parameter descriptions to include format/constraint details. Examples: 'organization: Filter by organization name or slug (e.g., "insee" or "ign"). Partial matches supported.' 'page_size: Number of results per page. Must be between 1 and 100. Default 20.' 'z: Zoom level as integer from 0 (world) to 20 (street detail).'
Add enums or pattern constraints to schema for enumerated parameters: 'resource' in calculate_route (valid values: 'bdtopo-valhalla', 'bdtopo-osrm', 'bdtopo-pgr'), 'profile' (valid: 'car', 'pedestrian'), 'cost_type' (valid: 'time', 'distance'), 'format' in export_cached_data (valid: 'json', 'geojson', 'csv').
Expand terse tool descriptions to 100-150 chars. Add context on when to use: 'list_wmts_layers: Retrieve all pre-rendered WMTS map tile layers available from IGN. Use this to discover available layers before calling get_wmts_tile_url. Each layer has a predefined zoom range and geographic coverage.' Current: 'Lister toutes les couches cartographiques WMTS disponibles (tuiles pré-générées)'.
Add pagination control parameters (page or offset) and document the response structure (total_count, has_next, next_cursor) for all search/list tools.
Score history
Overall score trend
↑ 49 points across a rubric change (v1 → v2)
54/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
D
54
2026-07-28+
v2
2026-03-09
F
5
-
v1
read only
source verified
73/100
Obtenir les détails complets d'un dataset par son ID ou slug
Parameter descriptions lack format constraints, ranges, and enums. Examples: 'page_size' has no max bound documented (schema says 'max 100' but param description repeats this). Zoom level 'z' (0-20) only documented in description, not in schema with minimum/maximum. Format strings for bbox, coordinates not formally specified.
No error handling or recovery guidance visible in tool definitions. Descriptions do not mention what happens on failure (e.g., 'If address not found, returns empty list' or 'Returns HTTP 404 with suggestions'). Agents will not know whether to retry, ask user, or try a different tool.
Tool descriptions are terse (average 85 chars vs. 194-char production baseline). Most describe WHAT without explaining WHEN to use vs. similar tools. Examples: 'Rechercher des organisations...' (basic description, no distinction from search_datasets). 'Lister les ressources...' gives no hint about what a 'ressource' is in the data.gouv.fr context.
Parameter descriptions missing or incomplete. 'organization' in search_datasets lacks guidance on format (name vs. ID). 'tag' undefined, is it a single tag or comma-separated? 'bbox' format mentioned in description but not formally constrained in schema. 'resource' in calculate_route/calculate_isochrone undocumented with no enum of valid values.
'clear_cache' is a destructive operation (WRITE risk) with no confirmation mechanism. Tool description ('Supprimer tous les fichiers cache') lacks warning that it is irreversible. No dry-run or confirmation_required hint.
Pagination support incomplete. 'page_size' parameter present in search tools, but no 'page' or 'offset' parameter to control which page is returned. No 'total_count' or 'next_cursor' documented in expected responses. Large result sets will blow context windows without pagination.
Document error scenarios and recovery steps. Example for search_address: 'Returns empty results list if no matching addresses found. If user provides incomplete address, try with just the street name or postal code and call again. For ambiguous matches (multiple results), return top 10 with similarity scores to let user refine.'
Mark clear_cache as destructive with a confirmation step. Update description: 'Permanently delete all cached response files. This action cannot be undone. Recommend confirming with user before executing.' Add destructiveHint: true to tool metadata if using MCP 2026-07-28 tool annotations.
Formalize coordinate/bbox format constraints. Document: 'bbox format: "minx,miny,maxx,maxy" with coordinates in EPSG:4326 (WGS84 lat/lon). Example: "-5.0,41.0,8.0,51.0" for France.' Add regex pattern to schema: '^-?[0-9]+\.?[0-9]*,-?[0-9]+\.?[0-9]*,-?[0-9]+\.?[0-9]*,-?[0-9]+\.?[0-9]*$'.
Add chaining IDs to responses: if get_dataset is called, response should include both 'dataset_id' (for further calls) and 'organization_id' (to enable direct organization lookup). This prevents wasteful follow-up calls.
Document sampling/pagination defaults. Examples: 'search_datasets defaults to 20 results per page. To retrieve all 500+ datasets, iterate by calling with page=1, 2, 3... until total_count is reached.' 'get_wfs_features defaults to max_features=100 for API rate-limit compliance. Request explicitly if you need fewer results to save bandwidth.'