MCP Shared Workspace
Starts an HTTP MCP server and exposes it publicly over HTTPS via Tailscale Funnel, printing a device/tailnet-bound hostname plus a generated private path that acts as the connector URL's access secret.
MCP Shared Workspace
A small, self-hosted Model Context Protocol (MCP) server for sharing project progress and UTF-8 text files between your own AI accounts.
The server stores everything on your computer. It does not call an AI API and does not require an API key. Your AI provider may still apply its normal account, subscription, connector, or usage limits.
What it provides
Read and update a shared project status.
List, search, and read files in a local shared_files folder.
Read a bounded section of a file to reduce unnecessary context.
Make guarded text patches using a file hash and expected match count.
Write a new text file or replace an existing one with a local backup.
Files written by the server are UTF-8 text files up to 500 KB. It does not edit binary assets.
Related MCP server: json-md-studio
Quick setup on Windows
Install Python 3.10 or newer, Tailscale, and Git.
Download this repository from GitHub or clone it.
Open PowerShell in the downloaded folder and run:
Set-ExecutionPolicy -Scope Process Bypass .\start_local.ps1The first run creates a local Python environment and installs the MCP dependency.
Put project files in shared_files. The server keeps status in data/state.json. The starter instructions at shared_files/WORKFLOW.md are versioned; your other project files are excluded from Git.
For local MCP clients, configure the client to launch the Python executable in .venv/Scripts/python.exe with server.py as its argument.
Connect Claude accounts through Tailscale Funnel
Use this when the MCP server should be reachable by Claude's website from multiple accounts. The computer owner signs into their own Tailscale account; each Claude account connects to the owner's server URL.
Sign in to Tailscale on this computer and enable Funnel for this device when Tailscale asks.
Run PowerShell as Administrator in this repository folder.
Start the server and Funnel for this session:
Set-ExecutionPolicy -Scope Process Bypass .\start_tailscale_mcp.ps1Copy the HTTPS hostname and the private MCP path printed in the PowerShell window. The complete URL is the hostname followed by that path.
In each supported Claude account, add the same complete URL under Settings -> Connectors -> Add custom connector. Enable the connector in each chat where you want to use it.
The hostname is tied to the owner's Tailscale device and tailnet. Keep this PowerShell window and computer on while using the connector. Press Ctrl+C to stop it. The next time you start it, the hostname and generated private path stay the same unless you change the device/tailnet or delete .mcp-secret.
Important access warning
The random path is the server's access secret. Anyone who gets the complete connector URL can read and change the shared status and files. Share it only with people you trust to have full read and write access. Do not publish it, paste it in issues, or commit it.
Never put passwords, API keys, personal account information, or private project files in the public Git repository. This repository contains only the reusable server and setup instructions. The data folder, your private files in shared_files, virtual environment, and .mcp-secret are ignored by Git; only the generic starter workflow in shared_files is tracked.
This setup has no hosted server cost, but it depends on your computer, internet connection, Tailscale availability, and the current limits/terms of the AI and tunnel providers. It does not transfer chat history or bypass provider usage limits; it shares only the data saved through this MCP server.
Publish your own public copy
On GitHub, create a new repository named mcp-shared-workspace and choose Public. Leave the options to add a README, .gitignore, or license unchecked because this folder already has them.
Open PowerShell in this folder and run these commands. Replace YOUR_USERNAME with your GitHub username:
git init git add . git status --short git commit -m "Initial public template" git branch -M main git remote add origin https://github.com/YOUR_USERNAME/mcp-shared-workspace.git git push -u origin main
Review the git status output before committing. It should contain only the reusable server, scripts, docs, license, and generic starter workflow. Do not use git add -f for ignored files.
Repository layout
server.py: MCP tools and local JSON/text persistence.
start_local.ps1: starts the local stdio server.
start_tailscale_mcp.ps1: starts the HTTP server and Tailscale Funnel on demand.
data/: local project state; excluded from Git.
shared_files/WORKFLOW.md: generic starter instructions available through MCP. Other files placed in shared_files/ are excluded from Git.
WORKFLOW.md: maintainer note explaining the MCP-visible workflow location.
.mcp-secret: generated URL path secret; excluded from Git.
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
A real-time, multiplayer filesystem for agents and humans: shared files, live edits, history.
Files and tables your AI reads and writes from any chat, with version history and sharing.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Project memory, documentation and controlled secret access for AI coding agents.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA secure file server for AI assistants that provides comprehensive file operations and text manipulation with configurable access levels and multiple connection modes.6MIT
- AlicenseNot gradedqualityCmaintenanceA local MCP server that enables AI agents to securely read, write, search, and edit JSON and Markdown files through a gated permission-scoped interface with configurable access modes.MIT
- AlicenseNot gradedqualityCmaintenanceA self-hosted remote MCP server for cross-model AI session handoff, allowing users to start a project in one AI assistant and continue in another via shared project logs, files with version history, and full-text search, all stored in a single SQLite file on their own server.5 npmMIT
- AlicenseNot gradedqualityAmaintenanceThis self-hosted MCP server connects an MCP client such as ChatGPT to approved local project folders, enabling file reading and editing, code search, shell command execution, and Git worktree workflows without uploading code to a third party. It also supports mounting external stdio or Streamable HTTP MCP servers as dynamic capabilities.603 npmMIT