ChatGPT2LocalBridge
Self-hosted MCP server that enables ChatGPT to access and operate on local project files after OAuth authorization, without uploading the workspace elsewhere.
README
<div align="center"> <img src="./docs/assets/logo.png" alt="ChatGPT2LocalBridge logo" width="180" /> <h1>ChatGPT2LocalBridge</h1> <p><strong>OAuth MCP Connector for approved local files.</strong></p> <p> <img alt="License" src="https://img.shields.io/badge/license-MIT-blue.svg" /> <img alt="Node" src="https://img.shields.io/badge/node-%3E%3D20-339933.svg" /> <img alt="MCP" src="https://img.shields.io/badge/MCP-Streamable%20HTTP-1769e0.svg" /> <img alt="Status" src="https://img.shields.io/badge/status-alpha-f59e0b.svg" /> </p> <p> <a href="./docs/index.html">Pages source</a> · <a href="./docs/showcase.html">Showcase</a> · <a href="./docs/human-tutorial.html">Human tutorial</a> · <a href="./docs/agent-computer-use.html">Agent tutorial</a> · <a href="./docs/alternatives.md">Alternatives</a> · <a href="./docs/sync-flows.md">Sync flows</a> </p> </div>

<!-- showcase:start --> <p align="center"> <img src="./docs/assets/thumbnail-responsive.png" alt="ChatGPT2LocalBridge responsive preview" width="900" /> </p> <!-- showcase:end -->
ChatGPT2LocalBridge is a self-hosted MCP server that lets ChatGPT access approved local workspaces after OAuth authorization. It is designed for people who want ChatGPT to inspect or operate on local project files without uploading the whole workspace elsewhere.
It is not a legacy ChatGPT plugin. It is best described as:
- MCP Server
- ChatGPT Custom Connector
- OAuth Local Workspace Bridge
Unofficial project. Not affiliated with OpenAI.
Route
ChatGPT
-> OAuth MCP Connector
-> HTTPS tunnel
-> http://127.0.0.1:3838/mcp
-> ChatGPT2LocalBridge
-> approved local workspace roots
ChatGPT does not directly mount your disk. It calls MCP tools, and every file operation is checked against bridge.policy.json.
30-Second Install
npx github:harzva/chatgpt2localbridge init --root ~/Projects
set -a; source .env.local; set +a
npx github:harzva/chatgpt2localbridge --http 3838
Local clone flow:
git clone https://github.com/harzva/chatgpt2localbridge.git
cd chatgpt2localbridge
npm install
npm run build
node dist/index.js init --root ~/Projects
set -a; source .env.local; set +a
node dist/index.js --http 3838
Health check:
curl -sS http://127.0.0.1:3838/health
Local operator console:
http://127.0.0.1:3838/app
ChatGPT Connector Setup
Expose the local server through HTTPS:
ngrok http 3838 --url=your-fixed-domain.ngrok-free.dev
Then create a ChatGPT Custom Connector:
| Field | Value |
|---|---|
| Name | ChatGPT2LocalBridge |
| URL | https://your-fixed-domain.ngrok-free.dev/mcp |
| Auth | OAuth |
When the authorization page opens, enter the unlock code from .env.local. Do not paste unlock codes or tokens into public chats, issues, screenshots, or commits.
Screenshot Walkthrough
| Step | Preview |
|---|---|
| Initialize local policy | ![]() |
| Run local MCP server | ![]() |
Check /health |
![]() |
| Create connector | ![]() |
| Authorize | ![]() |
| Test file listing | ![]() |
Full guides:
- Human setup tutorial
- Agent + Computer Use tutorial
- Visual showcase gallery
- Markdown human tutorial
- Markdown agent tutorial
Main MCP Tools
| Area | Examples |
|---|---|
| Project | project.snapshot, project.index, project.scripts |
| Code | code.read, code.read_range, code.search |
| Files | file.list, file.stat, file.write, file.patch, file.delete |
| Shell/tests | shell.exec, test.detect, test.run |
| Git | git.status, git.diff, git.checkpoint, git.revert |
| Runtime | workspace.*, task.*, process.*, port.check |
| Cloud sync | cloud.download |
| Bridge | bridge.status, bridge.health, bridge.logs, bridge.activity, service.restart |
File Sync And Activity
- Local files can be read by ChatGPT through approved MCP tools.
- ChatGPT/App-provided cloud file download URLs can be written back to local disk with
cloud.download. - Tool calls are persisted to
tool-calls.jsonl. - File writes, downloads, tasks, processes, and service restarts are persisted to
audit.jsonl. - The local console at
/appshows status, tool calls, and audit events.
See file sync flows.
Security Defaults
- Do not run unauthenticated on a public URL.
- Keep
allowedProjectRootsnarrow. - Never commit
.env.local,bridge.policy.json, OAuth stores, tokens, cookies, or unlock codes. - Prefer OAuth over URL tokens.
- Set
LOCALBRIDGE_DASHBOARD_TOKENbefore using/app. - Review shell deny rules before enabling shell access for broad workspaces.
See security model.
Alternatives
OAuth + fixed HTTPS tunnel is the default because it fits ChatGPT Custom Connectors well. Other options exist:
- OpenAI Secure MCP Tunnel, when available to your workspace
- Cloudflare Tunnel
- VPS reverse proxy
- Static bearer token for private clients
- Loopback-only no-auth testing
See alternatives.
GitHub Pages
The static product site lives in docs/. The repository includes a GitHub Actions workflow that deploys it to GitHub Pages after pushing to main.
Development
npm install
npm run typecheck
npm test
npm pack --dry-run
Render README and docs assets:
npm run docs:assets
npm run docs:preview
Public Release Checklist
- [ ] Enable GitHub Pages with the included workflow.
- [ ] Confirm
npm testpasses in GitHub Actions. - [ ] Keep
.env.localandbridge.policy.jsonuntracked. - [ ] Verify the ChatGPT connector uses OAuth and the correct
/mcpURL.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。





