PlaceRoot is an MCP server that connects AI agents to Overture Maps open map data to answer spatial questions, such as what is nearby, what is in a neighborhood, and how to get somewhere. It requires no API key, account registration, or vendor platform, and is an independent project of vibemapper.
Main capabilities and features:
- Returns sorted, compact results rather than raw GeoJSON data dumps, suitable for agents' context windows.
- Provides real routing: route and isochrone are based on actual street graph walking built from Overture transportation segments, covering the globe, rather than straight-line distance estimates.
- Supports reachability-filtered search: find_places(..., within={"minutes": 15, "mode": "walk"}) keeps only results that truly fall within walking, cycling, or driving reach on the street graph, rather than guessing with a radius or intersecting polygons yourself.
- Place data is rich and filterable, including categories, brands, confidence, operating status, contactability, and more, sourced from the Overture open dataset (contributed by Meta, Uber, TomTom, and others).
- Precise boundaries: you can search within the real administrative polygon of a named place, rather than a guessed circular radius.
- Zero configuration: reads Overture's public data directly, with nothing else to install besides the server itself.
How to use: It can be run directly from PyPI or npm without an installation step. Add the configuration to claude_desktop_config.json, or when using npm configure "command": "npx", "args": ["placeroot"]. Then you can ask the agent spatial questions, such as "What is around downtown Palo Alto", "Compare the bike shops in these two neighborhoods", "Plan my errand route: pharmacy, hardware store, post office".
Tools and configuration: There are 48 tools in total, divided into four series, with each answer kept within a few thousand tokens. It also provides seven workflow prompts (site selection, neighborhood comparison, errand planning, whether it is suitable for living, get to know my city, verify a listing description, plan an area visit) and three attachable resources; through the PLACEROOT_TOOLS setting you can load only the tool configuration you need, reducing schema overhead by up to 95%. The complete tool catalog, prompts, resources, andConfiguration is in docs/REFERENCE.md.
Use cases: Suitable for AI agents that need spatial reasoning, such as local life services, site selection analysis, neighborhood comparison, and trip and errand planning.
Limitations: Open data has clear boundaries—no real-time traffic, no opening hours, no ratings or photos. These are the parts PlaceRoot intentionally does not do.
Technical details: It is the only keyless MCP server that performs real graph path planning on globally open map data; every tool declares appropriate MCP annotations so clients know which calls are read-only before prompting the user. Stable GERS IDs let agents keep places across turns; local caching makes repeated queries return in milliseconds and work offline; the entire project can be self-hosted end to end.
Benchmarks against Mapbox MCP and Google Maps MCP, along with token efficiency data, are in the project documentation.
Leisure place coverage: Overture's places theme comes from business listings, so it is strong for businesses and thin for places families often visit on weekends (playgrounds, community parks, dog parks, beaches). These features are not missing; they are in another theme. Overture's base theme directly merges OpenStreetMap, and PlaceRoot already queries it in land_use_at and infrastructure_at, and the places tools also read it by default. No download, build, or hosting is needed—just one more live scan of the same public Overture release, which roughly expands playground coverage by 2.5x (in New York City, measured against the 2026-07-22.0 release, 1,552 versus 674; counts drift with releases, and 1,013 of them are more than 150 meters from any places-theme playground). All places tools answer from both at once, with no change to the tools, response structure, or category filters. The cost is scanning one more dataset per places query (also cached), and these rows have no confidence or business status and often no name—an unnamed playground returns name: null instead of being dropped. If latency matters more than coverage, the configuration can be adjusted. data_version reports this layer when the layer is active. Detailed explanation (including why live Overpass queries and raw OSM Parquet were tested and rejected) is in docs/RECREATION.md.
Development and documentation: Development setup, design rules, and the method for proposing tools are in CONTRIBUTING.md. Other documentation includes CHANGELOG, SECURITY, publishing, website, running a data mirror, and the leisure layer. Contact email: hello@placeroot.dev.
Privacy: PlaceRoot runs on your machine—no account, no API key, no sign-up. It has no telemetry and sends nothing to us; the only network traffic is your own queries going directly to the Overture Maps public data on AWS S3 (or a mirror you configure). Details are at placeroot.dev/privacy.html.
License and attribution: The code is MIT licensed. The queried data is the Overture Maps public release, licensed by theme—places uses CDLA-Permissive-2.0; OSM-derived themes (divisions, transportation, base) use ODbL, which requires attribution in any user-facing content you build: © Overture Maps Foundation · © OpenStreetMap contributors (ODbL). Theme-by-theme obligations are in docs/DATA-LICENSE.md.
MCP name: io.github.chuofringer/placeroot.
