rwee
Enables remote execution of Python/arcpy scripts on a machine with ArcGIS Pro, using a persistent worker process and file tools to run ArcGIS workflows and return results and artifacts.
Provides a prompt for answering questions about a DuckDB database using layer descriptions, and can return GeoJSON or GeoParquet for ArcGIS layers.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
mcp-rwee
An MCP server with four tools (read, write, edit, execute) and one prompt. A remote Claude session writes Python to a machine with ArcGIS Pro, runs it there, and gets the answer back.
Python files run in a long-lived worker process, so arcpy stays imported between runs, and a crash or timeout cannot stop the server.
Full manual: open docs/index.html. It has setup steps, diagrams, test results, security notes, troubleshooting and a complete worked example on a real project (section 15). Tested with ArcGIS Pro 3.7.3 and Python 3.13.13 on Windows 11 on Arm and on Windows on x64 hardware.
Related approach: Code Mode, by Philipp Schmid. The manual compares the two in section 2.2.
Quick start
On a Windows machine with ArcGIS Pro, in Command Prompt:
git clone https://github.com/mraad/mcp-rwee.git
cd mcp-rwee
uv venv --python "C:\Program Files\ArcGIS\Pro\bin\Python\envs\arcgispro-py3\python.exe" --system-site-packages
uv sync
mkdir .venv\conda-meta
mkdir C:\work
.venv\Scripts\python.exe -c "import secrets; print(secrets.token_urlsafe(32))"
set MCP_RWEE_TOKEN=<the value printed above>
.venv\Scripts\python.exe server.py --prompt prompts\arcpy.md --workdir C:\work --port 8765Before you start the server, edit the "Machine details" section of prompts\arcpy.md. The server listens on 127.0.0.1 only.
Related MCP server: AI Dev Assistant
Connect from another machine
The simplest way is an SSH tunnel. It needs SSH access to the Windows machine (manual section 7.6), but no certificate and no firewall rule for the server port.
On the client, open the tunnel and keep it open:
ssh -o ExitOnForwardFailure=yes -N -L 127.0.0.1:8765:127.0.0.1:8765 <user>@<host><user>is the Windows login name. KeepExitOnForwardFailure=yes. Without it, SSH can stay open with no tunnel if a different process uses the port, and the client then sends the token to that process.Make
mcp.json:{"mcpServers": {"rwee": {"type": "http", "url": "http://127.0.0.1:8765/mcp", "headers": {"Authorization": "Bearer ${MCP_RWEE_TOKEN}"}}}}Start Claude Code:
MCP_RWEE_TOKEN=<token> claude --mcp-config mcp.json
To start the server from the client, use manual section 7.10. To keep the server in Claude Code without --mcp-config, use section 7.11. That method writes the token as plain text in ~/.claude.json, and a session that is already open does not get the new tools.
To do all of this with one command from a Mac or Linux client, use remote.sh (manual section 7.12). It starts the server, opens the tunnel and registers rwee for this folder. stop removes all three. It shows each step. With -q, it shows only errors and warnings, and status still shows its result.
./remote.sh [-q] start [ssh-host] # host: argument or $RWEE_HOST; port: $RWEE_PORT or 8765
./remote.sh [-q] status [ssh-host]
./remote.sh [-q] stop [ssh-host]
./remote.sh [-q] fetch <download_path> <local file> [ssh-host]For a direct connection, use a TLS certificate and a firewall rule (manual sections 7.2 and 7.8). The server refuses a non-local --host without a certificate. The direct TLS path passed on the Parallels network, but not between two physical machines, and the firewall steps are not tested.
Artifacts
Code that execute runs can write files to the folder in RWEE_ARTIFACT_DIR, and the result lists them in artifacts. Clients list and read these files as read-only MCP resources, up to 12,000,000 bytes. A larger file downloads from download_path with the token. The server deletes run folders after --artifact-ttl (default 24h). See manual section 3.6.
Ask questions about a DuckDB database
prompts/northsea-duckdb.md answers questions about NorthSea.ddb, using the descriptions in Layers.json, and returns GeoJSON or GeoParquet for ArcGIS layers. It needs no arcpy, so it runs on a Mac:
uv run --with duckdb server.py --prompt prompts/northsea-duckdb.md --workdir <folder> --artifact-dir <folder> --port 8765See manual section 8.2.
Stop the server
Closing an SSH session does not stop a server that started in it, and its token keeps working. Find the process number, which is the last value on the line, and stop it:
netstat -ano | findstr LISTENING | findstr :8765
taskkill /F /T /PID <process number>Warning: execute is remote code execution by design. Anyone with the token runs code as the server's Windows user. Read section 12 of the manual before you expose the port.
Project structure
server.pycontains the FastMCP HTTP server, authentication, file tools, andScriptRunnerworker supervision.worker.pyexecutes Python scripts in a persistent process. Keep it standard-library-only so FastMCP dependencies do not load beside arcpy.tests/test_server.pycovers tools, prompts, worker behavior, authentication, and HTTP integration.prompts/arcpy.mdsupplies the task instructions.docs/index.htmlis the self-contained manual, including embedded images;README.mdis the quick start.remote.shmanages the remote server, SSH tunnel, and local Claude Code registration.
Development
Use Python 3.11 or newer and uv:
uv syncinstalls dependencies, including the pytest development group.uv run pytest -k test_nameselects matching tests.uv run server.py --prompt prompts/arcpy.md --workdir /path/to/workstarts locally; setMCP_RWEE_TOKENto at least 32 characters first../remote.sh status [ssh-host]checks remote operation.
There is no build step or configured formatter/linter. For arcpy, follow the Windows environment setup in the manual; uv run alone does not supply it.
Coding style
Follow existing Python style: four-space indentation, snake_case functions and variables, UPPER_CASE constants, and PascalCase classes. Reuse existing helpers and keep changes focused. Preserve exactly four tools (read, write, edit, execute) and one prompt (task).
Tests
On any machine (no arcpy):
uv run pytestOn Windows, after the quick start:
.venv\Scripts\python.exe -m pytest -qUse pytest with AnyIO's asyncio backend, test_* functions, and temporary-path fixtures. Add regression tests for behavior changes. No numeric coverage threshold is configured. Changes to server.py or worker.py also require Windows testing with real arcpy. Consult CLAUDE.md for platform constraints.
Commits and pull requests
History uses short imperative subjects, such as "Fix non-ASCII echo output in Windows shell commands"; merged subjects include PR numbers. Keep commits focused. PR descriptions should explain the behavior change, relevant issues, tests run, and platform limitations. Update the manual for user-visible changes and its embedded prompt copy when the prompt changes.
Security and configuration
Treat execute as code execution with the server user's privileges. Keep tokens out of commits and child environments. Retain loopback binding by default and TLS enforcement for non-local hosts; use SSH tunnels for remote access. The work directory is not a security sandbox.
License
MIT. See LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Work on your own computers from Claude or ChatGPT: run commands, edit and search files.
Use your Mac, Windows or Linux computer from ChatGPT, Claude or Codex: files, commands, documents.
Connect ChatGPT and Claude to your Mac, Windows or Linux with secure MCP remote access.
Connect ChatGPT, Claude, Cursor or VS Code to your Windows PC: files, terminal, browser, apps.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables AI tools like Claude to interact with a remote machine's file system and shell via a secure HTTPS endpoint. It provides standardized tools for executing shell commands, reading and writing files, and navigating directories.-
- FlicenseAqualityDmaintenanceBridges Claude Desktop to local Windows 10 development environments to enable repository reading, documentation searching, and isolated code execution. It provides a secure interface for running allowlisted terminal commands and scripts directly through natural language.4-
- FlicenseNot gradedqualityCmaintenanceEnables ISLI agents and MCP clients to dispatch natural-language coding and terminal tasks to a locally-installed Claude Code CLI, supporting both one-shot execution and persistent sessions with workspace and security controls.-
- AlicenseNot gradedqualityBmaintenanceEnables Claude to securely work on remote projects via SSH or WinRM, providing a restricted set of tools (read, patch, git, tests, build, docker, logs) limited to whitelisted directories.GPL 3.0