-
Notifications
You must be signed in to change notification settings - Fork 82
Description
Description
I am running a self-hosted instance of MetaMCP which aggregates approximately 10 different MCP servers. I am attempting to connect Code Mode to this MetaMCP instance to give the agent access to all aggregated tools at once.
I followed the official setup guide and the YouTube tutorial, configuring my .utcp_config.json to use the mcp call template with uvx mcp-proxy.
Current Behavior
The connection appears to initialize without crashing, but when the LLM (or manual execution) calls list_tools or search_tools, the result is always empty. No tools from the MetaMCP instance are discovered or registered by Code Mode.
Configuration
Here is the .utcp_config.json configuration I am using:
{
"manual_call_templates": [
{
"name": "metamcp",
"call_template_type": "mcp",
"config": {
"mcpServers": {
"metamcp": {
"command": "uvx",
"args": [
"mcp-proxy",
"--transport",
"streamablehttp",
"https://metamcp.myserver.com/metamcp/Default/sse"
],
"env": {
"API_ACCESS_TOKEN": "MYAPIACCESSTOKEN"
}
}
}
}
}
]
}Environment
- OS: [macOS ]
- Transport: SSE (Server-Sent Events)
- Proxy:
mcp-proxyviauvx - Aggregation: MetaMCP hosting ~10 servers
Expected Behavior
Code Mode should connect to the MetaMCP SSE endpoint, perform the handshake, and list_tools should populate with the tools aggregated by MetaMCP (e.g., github, filesystem, etc.), allowing the agent to call them via the metamcp namespace.
Possible Cause / Question
It seems like the authentication or tool discovery handshake is failing silently.
- Is
uvxcorrectly passing theenvvariables (specificallyAPI_ACCESS_TOKEN) to the spawnedmcp-proxyprocess? - Does Code Mode support tool discovery over an SSE connection bridged by
stdioin this specific configuration?
Any guidance on debugging the connection handshake would be appreciated.