MCP server for India's open geo data. Query layers, locate points, find nearby features, filter and group data. Also authors collect maps: register a share link, moderate contributions, and publish to the catalog.
bharatlas-mcp demonstrates solid tool design with 17 well-named, action-verb-prefixed tools (list_, get_, query_, locate, nearby, register_, moderate_, import_, edit_, publish, create_). All tools have descriptions (avg ~150 chars, within 10-1024 baseline). Input schemas are present and properly typed with JSON Schema. However, output schemas are NOT documented, critical gap for LLM planning. Parameter descriptions are present but inconsistent in depth. Error handling guidance is absent. No tool annotations (readOnlyHint/destructiveHint). Composition is strong: tools chain logically (list_layers → get_layer_schema → query_layer). Security concern: register_map accepts share_link with embedded tokens, should validate/sanitize. Overall: good naming and parameter coverage, but missing output documentation and error recovery patterns.
Create a new collect map headlessly (optional, requires COLLECT_API_KEY env var). Prefer creating online at collect.bharatlas.com. Returns three share links (admin, collect, view) with tokens in the URL #fragment.
Edit a collect map's settings: name, purpose, description, schema (geometry types and fields), category, and reference layer. Requires admin link.
Get full metadata for a layer: name, description, source, license, attribution, row count, geometry type, and direct download URLs for parquet, pmtiles, geojson, kml, and shapefile formats.
Get column names, types, and sample values for a layer. Use BEFORE querying to discover what columns exist and what values they contain. Essential for knowing the right column names for filters.
Get metadata for a registered collect map: name, purpose, description, schema (geometry types and fields), record counts by status, and moderation settings. Requires view, collect, or admin link.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does query_layer return? What fields are in each record?). Critical for multi-step agent workflows.
No error handling guidance. Tools lack recovery hints (e.g., 'Layer not found. Call list_layers() to discover valid IDs'). Agents cannot self-correct on failures.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes (moderate_record, import_points, edit_map, publish, create_map). Increases risk of unintended side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Query records in a registered collect map. Filter by status (pending, published, rejected), search by properties, or export all. Requires view, collect, or admin link.
Bulk import points into a collect map from a dataset you already hold. Requires admin link and a source URL + rights declaration.
List all available data categories (boundaries, city-wards, environment, water, agriculture, transport, infrastructure, health-edu, etc.) with counts of layers per category.
List available geo layers. Use to discover what data bharatlas has. Returns layer IDs, names, row counts, categories, and download formats. Filter by category (boundaries, city-wards, environment, water, agriculture, transport, etc.), admin level (state, district, block, village, etc.), source, or text search.
List all collect maps you have registered (via register_map). Shows map ID, name, status, and record counts.
List user-contributed datasets (community submissions). These work with query_layer and get_layer_schema exactly like curated layers. Filter by category, search by name/description, or sort by creation date or usefulness.
Find all admin boundaries and zones at a lat/lng. Returns state, district, block, village, and (for ~30 covered cities) the municipal ward layer containing the point. Use to answer 'what state/district/ward is this point in?'
Approve or reject a pending record in a collect map. Requires admin link.
Find features in a layer near a lat/lng within a distance radius. Returns features sorted by distance.
Submit a collect map's approved points to the bharatlas catalog FOR REVIEW. Nothing is auto-approved; a bharatlas maintainer approves the submission before it goes live. Once approved, it is queryable through list_submissions, query_layer, and locate. Requires admin link.
Query a layer's data with filters and grouping. Reads the parquet file at runtime. Supports: select specific columns, filter by column values, group by a column to get counts. Use get_layer_schema first to discover column names.
Register a collect.bharatlas.com share link to manage that map. The link's scope (view, collect, or admin) determines what you can do. View link: read/query/export published points. Collect link: same plus field capture. Admin link: everything including edit, moderate, import, and publish.
register_map accepts share_link with embedded tokens in URL fragment. No validation or sanitization visible. Tokens could leak into logs or agent traces. Should validate format and warn about token exposure.
Pagination parameters (limit, offset) present in list_* tools but no total_count or next_cursor returned. LLMs cannot determine if more results exist or how to fetch them efficiently.