Connect AgentDock from different clients
AgentDock connects to Claude Desktop, ChatGPT, Claude Code, Cursor, VS Code, Codex, TRAE, and WorkBuddy over Streamable HTTP MCP. Every example on this page uses:
https://agentdock.example.com/mcp
Replace it with your own AgentDock MCP URL.
Menu names, plan restrictions, and workspace policies may change between client releases. This page uses common current labels. When an entry is not visible, search settings for MCP, Plugins, or Connectors.
Before connecting
For a public client or cloud service, confirm that:
- AgentDock is available through a public HTTPS domain.
- The MCP URL ends with
/mcp(for example,https://agentdock.example.com/mcp). - Authentication is configured: remote clients typically require OAuth browser authorization or a Bearer Token in the HTTP request header. For full environment variable setup, see OAuth configuration.
- The reverse proxy forwards
/mcp,/register,/oauth/*, and/.well-known/*.
A local client running on the same computer as AgentDock may also connect to:
http://127.0.0.1:8765/mcp
A cloud client cannot reach 127.0.0.1 on your computer or server.
Keep real Bearer Tokens and OAuth credentials out of shell history and version-controlled workspace files. Use the client's secret-management capability or a restricted environment variable when available.
Claude Desktop
Availability of custom MCP connections and the number of servers you can add depend on the current plan and workspace policy.
-
Open Claude Desktop and go to Customize > Connectors.
-
Select +, then Add Connector.
-
Enter
AgentDockas the name. -
Enter this MCP Server URL:
https://agentdock.example.com/mcp -
Save and select Connect.
-
When the browser opens the AgentDock authorization page, enter
AGENTDOCK_OAUTH_PASSWORD.
ChatGPT
ChatGPT connects to AgentDock over Streamable HTTP MCP using a public HTTPS URL and OAuth browser authorization. Custom MCP plugins require Developer mode to be enabled in ChatGPT settings.
For the complete step-by-step walkthrough, screenshots, and troubleshooting, see Connect ChatGPT to AgentDock.
Claude Code
OAuth
Run:
claude mcp add --transport http agentdock https://agentdock.example.com/mcp
Then run this inside Claude Code:
/mcp
Select AgentDock and complete OAuth authorization in the browser.
Bearer Token
claude mcp add --transport http agentdock https://agentdock.example.com/mcp \
--header "Authorization: Bearer YOUR_TOKEN_HERE"
Cursor
- Open Cursor Settings.
- Select Tools & MCP.
- Select Add Custom MCP.
- Edit
.cursor/mcp.jsonin the project:
{
"mcpServers": {
"agentdock": {
"url": "https://agentdock.example.com/mcp"
}
}
}
Return to Tools & MCP, find AgentDock, select Connect, and complete OAuth in the browser.
For a Bearer Token:
{
"mcpServers": {
"agentdock": {
"url": "https://agentdock.example.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN_HERE"
}
}
}
}
VS Code
Create or edit .vscode/mcp.json in the workspace:
{
"servers": {
"agentdock": {
"type": "http",
"url": "https://agentdock.example.com/mcp"
}
}
}
You may also open the command palette:
Ctrl+Shift+P / Cmd+Shift+P
Run Add Server, choose HTTP (HTTP or Server-Sent Events), enter the AgentDock MCP URL and server ID, and choose a workspace or global configuration. Complete OAuth in the browser after saving.
For a Bearer Token:
{
"servers": {
"agentdock": {
"type": "http",
"url": "https://agentdock.example.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN_HERE"
}
}
}
}
Codex
Codex App
-
Open Codex and go to Settings > Plugins > MCP.
-
Select Add server.
-
Enter
AgentDockas the name. -
Choose Streamable HTTP as the transport.
-
Enter this MCP Server URL:
https://agentdock.example.com/mcp -
Save, select Authenticate, and complete OAuth in the browser.
Command line
codex mcp add agentdock --url https://agentdock.example.com/mcp
Follow the terminal prompt to complete OAuth login and authorization.
TRAE
- Open TRAE settings.
- Select MCP.
- Select Add > Add manually.
- Add this configuration:
{
"mcpServers": {
"agentdock": {
"url": "https://agentdock.example.com/mcp"
}
}
}
After saving, find AgentDock in the installed MCP servers, select Authenticate, and complete OAuth in the browser.
For a Bearer Token:
{
"mcpServers": {
"agentdock": {
"url": "https://agentdock.example.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN_HERE"
}
}
}
}
WorkBuddy
- Open WorkBuddy and select Skills on the left.
- Select MCP Servers on the right.
- Select Configure MCP.
- Add this configuration:
{
"mcpServers": {
"agentdock": {
"type": "http",
"url": "https://agentdock.example.com/mcp",
"disabled": false
}
}
}
After saving, return to the MCP list, find AgentDock, select Connect, and complete OAuth in the browser.
For a Bearer Token:
{
"mcpServers": {
"agentdock": {
"type": "http",
"url": "https://agentdock.example.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_TOKEN_HERE"
},
"disabled": false
}
}
}
Verify the connection
Do not stop after the client reports “connected.” Complete a real read-only call, for example:
Call AgentDock's agentdock_context and tell me the AgentDock version, operating system, path model, and available Skill and dynamic MCP capability indexes.
If the OAuth page does not open, verify:
curl -fsS https://agentdock.example.com/.well-known/oauth-authorization-server
curl -fsS https://agentdock.example.com/.well-known/oauth-protected-resource/mcp
A 302 from the authorization page usually means AgentDock is redirecting to the client as expected. If the browser does not continue, confirm that the reverse proxy preserves the Location header and that the browser or network policy does not block the callback URL.
See Configuration for complete authentication settings and Manual Linux deployment for public deployment.