Claude Local Bridge

Claude Local Bridge

A secure MCP server that allows Claude to read and write local files on your machine with explicit approval gating for each access.

Category
Visit Server

README

Claude Local Bridge

Access your local repos from Claude on your phone. No more being chained to your PC.

A secure bridge server that lets Claude (web, mobile, desktop) read and write files on your machine through an approval-gated MCP server. Every file access requires your explicit OK — nothing touches disk without you saying so.

┌──────────────────┐                          ┌────────────────────────┐
│  Claude Mobile   │    tunnel / LAN / VPN    │  Claude Local Bridge   │
│  Claude Web      │ ◄──────────────────────► │  (your PC)             │
│  Claude Desktop  │      MCP over SSE        │                        │
└──────────────────┘                          └──────────┬─────────────┘
                                                         │
                                               ┌─────────▼──────────┐
                                               │   Approval Gate    │
                                               │  (you approve each │
                                               │   file via dashboard│
                                               │   or phone)        │
                                               └─────────┬──────────┘
                                                         │
                                               ┌─────────▼──────────┐
                                               │   Local Files      │
                                               │  (sandboxed to     │
                                               │   workspace roots) │
                                               └────────────────────┘

Features

  • MCP Server — 7 tools Claude can call: browse, read, write, request access, list/revoke approvals, view audit log
  • Approval Gate — every file read/write blocks until you approve via the dashboard
  • Real-time Dashboard — industrial-themed web UI for managing approvals, browsing files, viewing audit logs
  • Sandboxed — only directories you whitelist are accessible
  • Token Auth — bearer token generated at startup for pairing
  • Audit Trail — every access logged with timestamp, action, path, and outcome
  • WebSocket — live push notifications for approval requests
  • Tunnel-ready — works with Tailscale, Cloudflare Tunnel, NetBird, or any reverse proxy

Quick Start

# Clone and install
git clone https://github.com/suhteevah/claude-local-bridge.git
cd claude-local-bridge
pip install -r requirements.txt

# Start the bridge (expose your project directories)
python -m app.main --roots ~/projects ~/code

The server starts and prints:

🔗 Bridge Ready
  HTTP API:   http://127.0.0.1:9120
  MCP (SSE):  http://127.0.0.1:9120/mcp/sse
  Dashboard:  http://127.0.0.1:9120/
  Token:      a1b2c3d4e5f6...

Connect Claude

Claude Desktop / Claude Code

Add to your MCP settings (claude_desktop_config.json or .claude/settings.json):

{
  "mcpServers": {
    "local-bridge": {
      "url": "http://localhost:9120/mcp/sse"
    }
  }
}

Claude Mobile (via tunnel)

Set up a tunnel (see tunnel.md) so your phone can reach the server, then use the tunnel URL as the MCP endpoint.

Quickest option (no account needed):

# In a second terminal
cloudflared tunnel --url http://localhost:9120
# Gives you: https://random-words.trycloudflare.com
# Use: https://random-words.trycloudflare.com/mcp/sse

MCP Tools

Tool Description
browse_files List the workspace file tree (no approval needed)
request_file_access Request approval to read/write — blocks until you decide
read_file Read a file (requires active READ approval)
write_file Write to a file (requires active WRITE approval)
list_approvals See all current approvals and their status
revoke_approval Revoke an existing approval
view_audit_log View the access history

How it works

  1. Claude calls browse_files to see your project structure
  2. Claude calls request_file_access with a reason ("I need to read src/main.py to fix the bug")
  3. Your dashboard shows the request — you approve or deny
  4. If approved, Claude can now read_file or write_file within the approved scope
  5. Everything is logged in the audit trail

Dashboard

The built-in web dashboard at http://localhost:9120/ lets you:

  • Approve/deny access requests in real-time (WebSocket-powered)
  • Browse your workspace file tree
  • View the full audit log with filtering
  • Monitor server status and active approvals
  • Revoke approvals at any time

Mobile-friendly design — manage approvals from your phone while Claude works.

HTTP API

The full REST API is also available for custom integrations:

Method Path Auth Description
GET /health No Server status and config
GET /files/tree Token Directory tree listing
GET /files/read?path=... Token Read file contents
PUT /files/write Token Write file contents
POST /approvals/request Token Request file access (blocking)
GET /approvals/ No List all approvals
GET /approvals/pending No List pending requests
POST /approvals/{id}/decide No Approve or deny
DELETE /approvals/{id} No Revoke approval
GET /audit/ No Recent audit entries
GET /audit/path?path=... No Audit for specific path
WS /ws/approvals No Real-time approval WebSocket

Tunnel Options

See tunnel.md for detailed setup guides. Summary:

Solution FOSS Cost Best For
Tailscale Client: BSD-3 / Headscale = full FOSS Free Personal phone access
Cloudflare Tunnel Client: Apache 2.0 Free Sharing with others, public URL
NetBird Full FOSS (BSD-3) Free self-hosted Full FOSS purity
FRP Apache 2.0 Free (needs VPS) Self-hosted everything

CLI Options

python -m app.main --roots DIR [DIR ...] [OPTIONS]

Options:
  --roots           Workspace directories to expose (required, multiple allowed)
  --host            Bind address (default: 127.0.0.1, use 0.0.0.0 for LAN/tunnel)
  --port            Port number (default: 9120)
  --max-file-size-mb  Max file size limit in MB (default: 10)

Security

  • Sandboxed roots — only directories you explicitly pass via --roots are accessible
  • Extension blocklist.env, .pem, .key, .p12, .pfx, .secret are always blocked
  • Path traversal prevention../ attacks are caught and rejected
  • Bearer token auth — API endpoints require the token generated at startup
  • Approval gating — every file read/write needs explicit human approval
  • Backup on write — automatic .bak file created before any overwrite
  • Audit logging — full trail of every access attempt (successful or not)

Project Structure

claude-local-bridge/
├── app/
│   ├── main.py              # Entry point, FastAPI + MCP setup
│   ├── mcp/
│   │   └── server.py        # MCP tools (browse, read, write, approve...)
│   ├── middleware/
│   │   └── auth.py          # Bearer token authentication
│   ├── models/
│   │   └── schemas.py       # Pydantic models and enums
│   ├── routers/
│   │   ├── approvals.py     # Approval CRUD endpoints
│   │   ├── audit.py         # Audit log endpoints
│   │   ├── files.py         # File read/write endpoints
│   │   └── ws.py            # WebSocket for real-time approvals
│   └── services/
│       ├── approval_service.py  # Approval state management
│       ├── audit_service.py     # In-memory audit log
│       └── file_service.py      # Sandboxed file operations
├── dashboard/
│   ├── index.html           # Dashboard shell
│   ├── style.css            # Industrial console theme
│   └── bridge.js            # API client + WebSocket
├── requirements.txt
├── tunnel.md                # Tunnel setup guide
└── README.md

License

MIT




Support This Project

If you find this project useful, consider buying me a coffee! Your support helps me keep building and sharing open-source tools.

ko-fi

Every donation, no matter how small, is greatly appreciated and motivates continued development. Thank you!

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured