Best for
- Geospatial analysts wanting a unified desktop GIS + AI workflow
- Developers who want to run local LLMs and connect custom MCP tools
- Open-source users comfortable building from source under GPL-3.0
A
Arion
Official
Not a standard server
The first agentic AI desktop app for geospatial analysis
This is not a typical connect-and-use MCP server; read the source repository before using it.
- Category
- Other #201 of 230
- Stars
- ★ 86 Popular
- Runtime
- Node.js 22+ · Prebuilt binary
- Credentials
- Optional API key
- License
- GPL-3.0
- Last commit
- ⚠
37FMRS · D
Arion is a distinctive early-stage open-source desktop app combining LLM chat, MCP tool integration, and map visualization for geospatial work. It is not an MCP server and cannot be configured as one in other MCP clients; with binary releases still pending, users must currently build from source.
Strongest · Maintenance 13/20 Weakest · Reliability 3/20
Why each score
- Reliability 3/20
- No manifest was cached, and the README contains no tool list, no MCP initialization/handshake evidence, and no tests or CI. More importantly, the README frames Arion as an Electron desktop app that 'supports user-defined MCP servers' — i.e., it appears to act as an MCP client, not a server exposing tools — which directly contradicts or at least leaves highly ambiguous the premise that this repo is itself 'an MCP server.' With no committed tests or CI workflows to substantiate any execution path, and per the static calibration cap (reliability ≤12 without verifiable execution evidence), the unclear object identity plus total absence of reproducibility evidence warrants a very low score.
- Security and permissions 7/20
- The README references credential storage via keytar, API keys for multiple LLM providers (OpenAI, Anthropic, Azure, Google, Ollama), local storage of chat history/knowledge base (SQLite/PGlite), a planned plugin system that would execute third-party code, and native GDAL binary invocation — all plausible credential and execution surfaces. However, there is no description of key encryption at rest, least-privilege boundaries, the scope of outbound network calls, or confirmation prompts before dangerous operations. No explicit red-line issue (hardcoded secrets, covert exfiltration) was found, but the incomplete disclosure of permission model, data flow, and confirmation mechanisms caps this at a middling score.
- Maintenance 13/20
- The project has a clear GPL-3.0 license, a CONTRIBUTING.md, changelogs, a support email, is not archived, and shows some community attention (83 stars, only 4 open issues). However, there is a notable version inconsistency: the text announces 'v0.0.14 Released' while linking to a changelog file named 'v0.3.9-2025-12-29.md,' undermining confidence in release governance. Statements like 'Binary Releases Coming Soon' and multiple 'Planned' features also indicate an early-stage, still-forming project.
- Documentation 9/20
- The README is reasonably detailed and actionable for installation and cross-platform builds (Windows/macOS/Linux), native module rebuilding, and GDAL fallback behavior. However, it is entirely missing MCP tool parameter documentation, authentication details, rate/cost limits, troubleshooting guidance, or example configuration for connecting this app as an MCP server to a mainstream client (e.g., Claude Desktop). The version-number/changelog-filename mismatch further signals documentation upkeep gaps.
- Setup experience 5/20
- The README's setup steps describe compiling a cross-platform desktop application locally, requiring platform-specific build toolchains, native module rebuilds, and GDAL installation — a nontrivial, environment-dependent process — and explicitly states that binary releases are 'coming soon,' meaning no ready-to-use installer currently exists. Critically, there is no example configuration showing how to register this as an MCP server in any mainstream MCP client, so evidence for 'MCP server setup ease' specifically is essentially absent.
Static review · not runListed 2026-09-07
Read the FMRS scoring method →Fit and risk
What it can accessReads local filesWrites / deletes local filesUses the networkConnects to a database
Not for
- Users looking for an MCP server entry to configure in clients like Claude Desktop (Arion is not an MCP server and ships no server manifest)
- Users who need official turn-key installers (binary releases are not yet published)
- Non-technical users unwilling to build an Electron app and compile native modules
Required permissions
- Local file system read/write for app data (SQLite) and the knowledge base (PGlite)
- Network access to call cloud LLM provider APIs
- Access to a local Ollama instance if local models are configured
- Execution of system GDAL binaries for raster processing
- Connections to user-defined MCP servers and execution of their tools
Risks and side effects
- The project is early-stage (v0.0.14); several features (plugin system, Python agent runtimes, SpatiaLite integration) are planned only
- Building from source is required; native modules (better-sqlite3, keytar) can fail to compile
- Connecting custom MCP servers executes external code whose safety you must evaluate yourself
- With cloud LLMs, query content is sent to third-party APIs
Setup
Before you start
Runtime：Node.js 22+ · Prebuilt binary
Requires Node.js 22+ and npm 8+. Clone the repository, run npm install, and start development mode with npm run dev. Production builds need platform toolchains: Windows requires Visual Studio Build Tools (npm run build:win), macOS requires Xcode Command Line Tools (npm run build:mac, produces a .dmg), Linux requires build-essential, python3, and libsecret-1-dev (npm run build:linux, produces AppImage/Snap/.deb). Raster processing uses GDAL: install via brew install gdal (macOS) or sudo apt install -y gdal-bin libgdal-dev (Ubuntu); Windows builds bundle GDAL and do not fall back to system GDAL. Official binary installers are not yet available.
Check that it works
Arion is an MCP client desktop app, not an MCP server; run npm run dev to launch it, then chat with an LLM and confirm map visualization works to verify the install.
Troubleshooting
- Verify Node.js 22+ and npm 8+ are installed
- On Windows build errors, install Visual Studio Build Tools (C++ workload)
- On macOS build errors, run `xcode-select --install`
- After dependency updates, run `npm run rebuild` to rebuild native modules
- Verify GDAL with `gdalinfo --version`; macOS/Linux fall back to system GDAL on PATH, Windows does not
Things to try
Once connected, you can ask your AI assistant things like:
- Add a buffer around a feature on the map and show the result
- Ingest this PDF document into the knowledge base for later retrieval
- Configure a local Ollama model to analyze geospatial data
- Connect my user-defined MCP server and call its tools
Use cases
Running geospatial queries and analysis through a chat interface
Visualizing and manipulating geospatial data on a MapLibre GL map
Ingesting PDF/DOCX/TXT documents into a local vector store for RAG
Using local LLMs via Ollama for geospatial tasks
Extending agent capabilities with custom MCP servers exposing Python/TypeScript tools
Overview
Arion is a cross-platform desktop application (Electron + React + TypeScript + Vite) by GeoRetina Inc. for advanced geospatial analysis and agentic workflows. It supports multiple LLM providers (OpenAI, Google Gemini, Azure, Anthropic, Ollama) via the Vercel AI SDK, can connect to user-defined MCP servers to access external Python/TypeScript tools, and includes dynamic map visualization with MapLibre GL plus a local RAG knowledge base built on PGlite with pgvector. Note: Arion is an MCP client / desktop app, not an MCP server; the repository provides no server manifest, tool list, or transport for serving MCP.
Similar servers
Hevy MCP
77 · B
Manage your Hevy workout data from AI assistants via MCP.
Firecrawl MCP Server
75 · B
Firecrawl's official MCP server for web search, scraping, and structured extraction for AI agents
BioMCP
75 · B
One binary. One grammar. Evidence from the biomedical sources you already trust.
Compartment
75 · B
Encrypted, fully offline long-term memory for AI agents
Source revision 680fb56e6197 Data synced 2026-10-19 Read the FMRS scoring method