chat-share
MCP server for publishing AI agent conversations as public links with optional password protection and expiration, also providing tools to revoke shared chats and retrieve link metadata.
README
chat-share
Share AI agent conversations via unique public links. Self-hosted, with a REST API + MCP server for agents, and an admin panel protected by OIDC login (via PocketID).
It's like ChatGPT's "Share" feature, but self-hosted and designed so AI agents themselves (Hermes, Claude, Codex, etc.) publish and manage the links.
✨ Features
- REST API + MCP for AI agents to publish conversations and get a
/s/<id>link. - Public link rendered as nice HTML (markdown, syntax highlighting, role-styled bubbles).
- Optional password (argon2-hashed) to protect the link.
- Optional expiration (
1h,24h,7d,30d, or never). - Admin panel with OIDC (PocketID) login: view all chats, view counts, expiration, revoke/delete links.
- View counter and status (active / expired / revoked).
- SQLite (WAL), no external dependencies. Small multi-arch Docker image.
🚀 Docker quickstart
docker run -d --name chat-share \
-p 3000:3000 \
-v chat-share-data:/app/data \
-e BASE_URL="https://share.yourdomain.cl" \
-e SESSION_SECRET="$(openssl rand -hex 32)" \
-e AGENT_API_KEY="a-secret-key" \
-e OIDC_ISSUER="https://id.yourdomain.cl" \
-e OIDC_CLIENT_ID="chat-share" \
-e ADMIN_ALLOWED_SUBS="<user-sub>" \
ghcr.io/<your-user>/chat-share:latest
Or with docker-compose.yml (see docker-compose.yml in the repo root):
services:
chat-share:
image: ghcr.io/<your-user>/chat-share:latest
restart: unless-stopped
ports: ["3000:3000"]
volumes: ["chat-share-data:/app/data"]
env_file: .env
volumes:
chat-share-data:
Copy .env.example to .env, fill it in, and run docker compose up -d.
⚙️ Environment variables
| Variable | Required | Default | Description |
|---|---|---|---|
PORT |
no | 3000 |
HTTP port |
BASE_URL |
yes (prod) | http://localhost:3000 |
Public base URL used to generate links |
DB_PATH |
no | ./data/chat-share.db |
SQLite path (mount a volume) |
AGENT_API_KEY |
yes | — | Agent keys (comma-separated) |
SESSION_SECRET |
yes | — | HMAC secret for cookies (≥32 random chars) |
OIDC_ISSUER |
yes* | — | PocketID issuer (e.g. https://id.yourdomain.cl) |
OIDC_CLIENT_ID |
yes* | — | Admin panel client ID |
OIDC_CLIENT_SECRET |
no | — | Client secret (if the client is confidential) |
OIDC_AUDIENCE |
no | — | Audience required when validating MCP OAuth access tokens |
ADMIN_ALLOWED_SUBS |
no | — | OIDC subs allowed in the admin panel (comma-separated) |
COOKIE_SECURE |
no | true |
Secure flag on cookies (set false only on local HTTP) |
BODY_LIMIT |
no | 1048576 |
Request body limit in bytes for API/MCP/unlock |
* OIDC is optional: without OIDC_ISSUER/OIDC_CLIENT_ID the admin panel is disabled,
but the API and public links keep working.
🤖 Publishing from an AI agent
Via REST API (curl)
curl -X POST https://share.yourdomain.cl/api/chats \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"title": "Network issue fix",
"agent": "hermes-agent",
"messages": [
{"role": "user", "content": "How do I configure DNS?"},
{"role": "assistant", "content": "Edit `/etc/resolv.conf`..."}
],
"password": "optional-123",
"expires_in": "7d"
}'
Response:
{
"id": "mWqq9mPEdrRZ",
"url": "https://share.yourdomain.cl/s/mWqq9mPEdrRZ",
"expires_at": "2026-08-09T19:39:52.728Z",
"password_protected": true
}
Via MCP
The /mcp endpoint is an MCP Streamable HTTP server (stateless). Tools:
| Tool | Description |
|---|---|
share_conversation |
Publishes a conversation → returns a URL. Asks the user about password/expiration if they didn't specify any. |
revoke_shared_chat |
Revokes a link (id). |
get_shared_chat_info |
Link metadata (views, expires_at, status). |
MCP authentication — two mechanisms are supported:
- API key (same as the REST API): header
Authorization: Bearer <AGENT_API_KEY>. - OAuth 2.0 against PocketID: the server acts as a resource server that delegates
authorization to PocketID. Standard discovery flow:
- Metadata:
https://share.yourdomain.cl/.well-known/oauth-protected-resource/mcp - The client does an OAuth login at PocketID and receives an access token sent as
Authorization: Bearer <access_token>. - The server validates the token against PocketID's JWKS (
OIDC_ISSUERconfig).
- Metadata:
Configure in Hermes (config.yaml or hermes mcp add):
mcp:
servers:
chat-share:
transport: streamable-http
url: https://share.yourdomain.cl/mcp
headers:
Authorization: "Bearer YOUR_API_KEY"
If you prefer OAuth over an API key, point the MCP client at the metadata URL above and Hermes/Claude/Codex will perform the OAuth login against PocketID automatically.
When the MCP authenticates via OAuth, the agent field of published chats reflects the
user's identity (PocketID email/sub) instead of a generic name.
🔐 Admin panel
- Create an OIDC client in PocketID with redirect URI
https://share.yourdomain.cl/admin/callback. - Configure
OIDC_ISSUERandOIDC_CLIENT_ID(plusOIDC_CLIENT_SECRETif applicable). - Go to
/admin. You'll see all shared chats: title, agent, date, expiration, views, password protection, status, and actions (copy URL / revoke).
The flow uses Authorization Code + PKCE, validates the id_token against the issuer's JWKS
and creates a signed stateless session (HMAC).
🐳 Docker image / CI
- Multi-stage Dockerfile → small runtime image (
oven/bun:1-alpine). - GitHub Actions (
.github/workflows/docker.yml): on every push tomainor tagv*it builds forlinux/amd64andlinux/arm64and publishes to GHCR (ghcr.io/<repo>/chat-share).
🧪 Development
bun install
bun run src/index.ts # requires SESSION_SECRET and AGENT_API_KEY
bunx tsc --noEmit # typecheck
bun test # tests
📄 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。