location-platform MCP server
Commit Oct 21, 2026Allows AI clients to query Google Maps Timeline history and live location data through self-hosted MCP servers.
Also known as blakepennel/location-platform
- Stars
- 1
- Tools
- 10
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 21, 2026
- License
- MIT
What is location-platform MCP?
location-platform allows AI clients to query Google Maps Timeline history and live location data through self-hosted MCP servers. It exposes 10 tools, such as where_was_i, time_at_place, visits_near.
Summary generated from the project's README
About location-platform MCP server
Querying your personal movement history and current position from inside an AI assistant is what location-platform enables. This self-hosted setup connects Claude or ChatGPT to your Google Maps Timeline and live location data through two distinct MCP servers. By running the system via Docker, you keep your sensitive location information on your own hardware rather than in the cloud.
The system uses a timeline-mcp server to access historical data like visits, trips, and activities, while a live-location-mcp server provides real-time updates through Google Maps Location Sharing. Users can control data precision with semantic, approximate, or exact settings. Security is handled through an OAuth 2.1 layer, ensuring Google credentials never leave your server. The setup includes automated synchronization and a persistent browser profile to manage re-authentication without constant manual intervention.
Written from the project's README. Every use case is checked against a quote from it.
location-platform MCP tools 10
Detected from the READMEwhere_was_itime_at_placevisits_neardistance_traveledtimeline_statuswhere_am_irecent_locationswhere_was_i_recentlymovement_sincelocation_status
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableblakepennel/location-platform is public and reachablegithub.com/blakepennel/location-platform ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-21github.com/blakepennel/location-platform ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model10 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub📍 location-platform
Ask your AI assistant where you've been — and where you are right now.
Self-hosted MCP servers that turn your Google Maps Timeline and live location into tools for Claude, ChatGPT, or any MCP client.
Why
Google stopped offering Timeline on the web, and your location history now lives only on your phone (plus an encrypted cloud backup). This project pulls that history back out, keeps it on your own server, and lets an LLM answer questions about it — alongside a separate live feed for "where am I right now?".
What you can ask
"Where was I last Tuesday around 3pm?" →
where_was_ifinds the visit or trip covering that moment.
"How much time did I spend at the gym in September?" →
search_places+time_at_placetotals every visit.
"Summarize my week." →
summarize_weekreturns days, top places, distance, and how you got around.
"When was the last time I went to that taco place downtown?" →
visit_historylists every visit, newest first.
"How far did I drive this month vs. walk?" →
distance_traveledbreaks it down by activity.
"Where am I right now, and have I moved since noon?" →
where_am_i+movement_sincefrom the live feed, with data age shown.
How it works
flowchart LR
subgraph google["Google"]
TB[(Timeline<br/>cloud backup)]
LS[(Maps Location<br/>Sharing)]
end
subgraph server["Your server (Docker)"]
TS[timeline-sync<br/><sub>Python · every 6 h</sub>]
TJ[/Timeline.json/]
TM[timeline-mcp<br/><sub>13 tools</sub>]
LP[live poller<br/><sub>every 60 s</sub>]
LD[(live.sqlite)]
LM[live-location-mcp<br/><sub>5 tools</sub>]
AUTH{{OAuth 2.1}}
end
TB --> TS --> TJ --> TM
LS --> LP --> LD --> LM
TM & LM --- AUTH
AUTH --> C[Claude]
AUTH --> G[ChatGPT]
History and live are deliberately two separate systems. Timeline is Google's semantic reconstruction (visits, trips, activities, Place IDs) and arrives hours late. Live Location Sharing is raw point observations, seconds old. They live in separate databases behind separate MCP servers and are never merged. The model picks the right tool, and every answer says how old its data is.
Features
| 🗺️ Full Timeline history | Syncs straight from Google's encrypted Timeline backup (no phone export, no Takeout) and follows every backup corpus, so nothing is cut off. |
| 🏷️ Real place names | Resolves Place IDs to names automatically after each sync. |
| 📡 Live location | Polls Google Maps Location Sharing through a dedicated recipient account. The session refreshes itself with cookie rotation and a persistent browser profile. |
| 🔁 Hands-off re-auth | When Google revokes the Timeline token, a persistent browser signs back in. The password is sealed by the host's TPM 2.0 (systemd-creds), so a stolen disk or backup can't use it. It stops and asks for a human on any CAPTCHA or unexpected challenge. |
| 🔐 Two auth layers | Google credentials never leave the server. MCP clients get only short-lived, audience-bound OAuth 2.1 tokens (RFC 9728 discovery, PKCE, JWKS verification). |
| 🎚️ Precision controls | Three levels with a per-server cap on what an LLM may see: semantic (place names only, no coordinates), approximate (~1 km), or exact. |
| 🧾 Logs that can't leak | Structured logs with automatic redaction of tokens, cookies, keys, and anything that looks like a coordinate. |
| 🧪 Synthetic by default | Every test and fixture uses invented coordinates. A leak scanner checks the repo for secrets and real locations. |
🐳 One docker compose up |
OAuth server, both MCP servers, the poller, the sync scheduler, and noVNC browser views. Ports bind to 127.0.0.1 only. |
| 💬 ChatGPT without exposure | Optional OpenAI Secure MCP Tunnel services: an outbound-only connection, with nothing published to the internet. |
Tools
| Tool | |
|---|---|
where_was_i |
A point in time |
visits · timeline_between |
A day or range |
summarize_day · summarize_week |
Digests |
search_places · visit_history |
Find a place |
time_at_place |
Total dwell time |
visits_near |
Radius search |
trips · activities |
Movement |
distance_traveled |
Distance by mode |
timeline_status |
Data freshness |
| Tool | |
|---|---|
where_am_i |
Latest fix |
recent_locations |
Points in a window |
where_was_i_recently |
Nearest fix to a time |
movement_since |
Distance and moves |
location_status |
Poller health |
Every tool is read-only and annotated as such. There is no SQL, file access, or bulk-export tool.
Quick start
Try it with synthetic data. No Google account needed.
git clone --recurse-submodules https://github.com/blakepennel/location-platform.git
cd location-platform
npm install
cd timeline-sync && python -m venv .venv && .venv/bin/pip install -e ".[dev]" && cd ..
cp .env.example .env
npm run seed # invented Timeline + live observations
npm run dev # OAuth :8700 · timeline-mcp :8701 · live-location-mcp :8702
Point an MCP client at http://localhost:8701/mcp or http://localhost:8702/mcp. It discovers
the OAuth server and opens a sign-in page. For Claude Code, the stdio transport skips OAuth
entirely; see DEVELOPMENT.md.
Running it for real (a home server, your own Google data):
docker compose up -d --build
DOCKER.md walks through the one-time Google steps, the noVNC browser logins, TPM setup for automatic re-auth, and connecting Claude and ChatGPT.
Project layout
location-platform/
├── timeline-sync/ Python · syncs + decrypts the Timeline backup (wraps arkenoi/timeline-export)
├── timeline-mcp/ TypeScript · historical MCP server over its own SQLite index
├── live-location-mcp/ TypeScript · live poller + MCP server
├── mcp-auth/ OAuth 2.1 resource-server verifier + local dev authorization server
├── shared/ logging/redaction, time/geo, sqlite, HTTP host
├── schemas/ export contract + status JSON schemas
├── tools/ dev runner, e2e harness, seed data, leak scanner, re-auth driver
└── docker/ container entrypoints (noVNC, sync loop, Google browser)
Testing
npm test # 246 Node + ~125 Python tests
npm run test:e2e # full stack: OAuth + both servers + a real MCP client
npm run leak-scan # secrets and real-coordinate scan
Docs
| ARCHITECTURE.md | Diagrams and data flow |
| DOCKER.md | Deploying on a server |
| DEVELOPMENT.md | Local setup and real Google onboarding |
| AUTH.md | The two authentication layers |
| SECURITY.md · THREAT_MODEL.md | What's protected, and from whom |
| Per-project READMEs | timeline-sync · timeline-mcp · live-location-mcp |
Disclaimer
This is a personal project that reads your own data through undocumented Google endpoints: the Timeline cloud backup via arkenoi/timeline-export (a pinned submodule under its own MIT license), and Maps Location Sharing. Google can change or block them at any time, and using them may conflict with Google's Terms of Service. It is not affiliated with or endorsed by Google. Use it only with accounts you own, at your own risk.
License
location-platform MCP questions
What tools does location-platform MCP provide?
It exposes 10 tools, including where_was_i, time_at_place, visits_near, distance_traveled, timeline_status.
Is my Google data shared with the AI provider?
No, Google credentials never leave your server; MCP clients only receive short-lived, audience-bound OAuth 2.1 tokens.
Can I control how much location precision the AI sees?
Yes, there are three precision levels available: semantic (place names only), approximate (~1 km), or exact.
How does the server handle Google account re-authentication?
The system uses a persistent browser that signs back in when tokens are revoked, with passwords secured by the host's TPM 2.0.