chatgpt-context-mcp
Imports private ChatGPT conversations into MCP clients like Codex and Claude Code. Provides tools to fetch remote conversation content and serve cached context via local stdio.
README
ChatGPT Context MCP
A local stdio MCP server for importing private ChatGPT conversation context into Codex, Claude Code, and other MCP clients.
Disclaimer
This is an unofficial project. It is not maintained, endorsed, or supported by OpenAI.
This tool uses private, non-public ChatGPT web backend APIs. The behavior was inferred from web traffic and may change, break, or return incomplete data at any time. This project is provided only for technical research, personal context migration, and local MCP experimentation.
No guarantees are made about correctness, reliability, stability, completeness, legality, compliance, or fitness for any purpose. You are responsible for your own use, including token handling, data privacy, local laws, organization policy, and relevant service terms. Bearer tokens are sensitive temporary credentials and should be handled like passwords. Use at your own risk.
Features
- Imports private ChatGPT conversation URLs into MCP clients.
- Refreshes remote conversation content by default before returning context.
- Caches JSON, Markdown, messages, and metadata locally.
- Supports cache-only reads for previously imported conversations.
- Provides MCP tools and cache-backed resources.
- Documents setup for Codex and Claude Code.
Supported Scope
Supported:
https://chatgpt.com/c/{conversation_id}
https://chat.openai.com/c/{conversation_id}
Not supported:
- Shared links such as
https://chatgpt.com/share/{share_id} - ChatGPT Projects
- Bulk history export
- Attachment, image, or Canvas downloads
- Browser automation or cookie extraction
How It Works
import_chatgpt_url parses a private ChatGPT conversation URL, fetches the full remote conversation through ChatGPT's web backend API, compares a stable content hash with local metadata, updates the cache when content changed, and returns bounded Markdown/messages/JSON to the MCP client.
get_chatgpt_context reads local cache only. It does not check the remote conversation.
Install From Source
Requirements:
- Node.js 20+
- npm
git clone https://github.com/OWNER/chatgpt-context-mcp.git
cd chatgpt-context-mcp
npm install
npm run build
Run checks:
npm run check
Token Setup
Get a fresh ChatGPT web bearer token from your browser DevTools. Do not paste tokens into agent chat or commit them to config files.
You can provide credentials through the current process environment:
export CHATGPT_BEARER_TOKEN="<chatgpt-bearer-token>"
export CHATGPT_ACCOUNT_ID=""
For desktop apps, use the wrapper script with a private fallback env file:
mkdir -p "$HOME/.config/chatgpt-context-mcp"
chmod 700 "$HOME/.config/chatgpt-context-mcp"
cat > "$HOME/.config/chatgpt-context-mcp/env" <<'EOF'
export CHATGPT_BEARER_TOKEN="<chatgpt-bearer-token>"
export CHATGPT_ACCOUNT_ID=""
EOF
chmod 600 "$HOME/.config/chatgpt-context-mcp/env"
The wrapper uses this precedence:
- Existing process environment variables.
- Values from
$HOME/.config/chatgpt-context-mcp/env.
Set CHATGPT_CODEX_MCP_ENV_FILE to point at a different env file.
Codex Configuration
Built mode:
[mcp_servers.chatgpt_context]
command = "<repo-path>/scripts/start-mcp.sh"
args = []
env_vars = ["CHATGPT_BEARER_TOKEN", "CHATGPT_ACCOUNT_ID", "CHATGPT_CODEX_MCP_CACHE_DIR", "CHATGPT_CODEX_MCP_ENV_FILE"]
startup_timeout_sec = 10
tool_timeout_sec = 120
default_tools_approval_mode = "auto"
Development mode:
[mcp_servers.chatgpt_context]
command = "npm"
args = ["run", "start:mcp", "--prefix", "<repo-path>"]
env_vars = ["CHATGPT_BEARER_TOKEN", "CHATGPT_ACCOUNT_ID", "CHATGPT_CODEX_MCP_CACHE_DIR"]
startup_timeout_sec = 10
tool_timeout_sec = 120
default_tools_approval_mode = "auto"
Restart Codex or open a new session after changing MCP configuration.
Claude Code Configuration
Configure this as a stdio MCP server:
{
"mcpServers": {
"chatgpt_context": {
"command": "<repo-path>/scripts/start-mcp.sh",
"args": [],
"env": {
"CHATGPT_CODEX_MCP_ENV_FILE": "$HOME/.config/chatgpt-context-mcp/env"
}
}
}
}
Client-specific UI and config locations may vary. The server itself is a standard stdio MCP server.
MCP Tools
import_chatgpt_url: Fetches a private ChatGPT conversation URL from the remote API by default, updates cache when content changed, and returns bounded context.get_chatgpt_context: Reads a cached conversation. Does not access the network.list_chatgpt_imports: Lists locally cached imports.verify_chatgpt_auth: Checks token presence, expiration, account id detection, and lightweight API access.
MCP Resources
Cache-backed resources:
chatgpt://conversation/{conversation_id}/metadata
chatgpt://conversation/{conversation_id}/markdown
chatgpt://conversation/{conversation_id}/messages
chatgpt://conversation/{conversation_id}/json
Resource reads are cache-only and do not refresh remote content.
Cache and Privacy
Imported conversations are cached as plaintext under:
$HOME/.cache/chatgpt-context-mcp/
Override the cache location:
export CHATGPT_CODEX_MCP_CACHE_DIR="/path/to/cache"
Cache files may contain private prompts, assistant replies, code, and other sensitive content. The server does not write bearer tokens to disk.
Troubleshooting
If verify_chatgpt_auth fails:
- Confirm
CHATGPT_BEARER_TOKENis available to the MCP server. - Refresh the bearer token if it expired.
- For Team or Enterprise accounts, set
CHATGPT_ACCOUNT_IDif auto-detection fails. - Restart the MCP client after changing config or environment variables.
If import_chatgpt_url returns UNSUPPORTED_URL, use a private /c/{conversation_id} URL rather than a /share/{share_id} URL.
If remote access fails but cache exists, the error includes cache_available: true. Use get_chatgpt_context only when stale cached context is acceptable.
Development
npm install
npm test
npm run typecheck
npm run build
npm run check
Run the server locally:
npm run start:mcp
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。