Setup 1:
I have Claude desktop setup in windows and using the mcp-proxy.exe tool (installed on windows) able to call this MCP which is running in docker in another machine. Here is the sample config. And this works fine and is using the Claude Sonnet 4 model
{
"mcpServers": {
"remoteMCP": {
"command": "C:\\Users\\Manju\\Downloads\\mcp-proxy\\mcp-proxy.exe",
"args": ["http://10.173.253.48:8443/sse"]
}
}
}
Setup 2:
I have liteLLM and anythingLLM running as docker container and using the Gemini-2.0-flash model. They are also configured with this MCP server. In both the scenarios, the functions / tools hosted by this MCP server is visible in both liteLLM and anythingLLM. However, when chatting both complain about the schema problem. So looks like this MCP server does not confirm to the stricter naming convention of Gemini models or more specifically VertexAI?
The config in anytingLLM
{
"mcpServers": {
"kube": {
"url": "http://10.173.253.48:8443/sse",
"type": "sse",
"args": [],
"env": { // optional, some MCP servers may require additional environment variables
"MY_ENV_VAR": "my-env-var-value"
}
"anythingllm": {
"autoStart": false
}
}
}
}
This is what I found out from logs:
Logs from anythingLLM
[backend] info: [AgentLLM - gemini-2.0-flash] Invalid function tool call: Invalid argument schema match..
[backend] info: [AgentLLM - gemini-2.0-flash] Will assume chat completion without tool call inputs.
❌ Non-Compliant Function Names
These function names violate VertexAI's naming rules because they contain colons (:), which are not allowed:
remoteMCP:configuration_view
remoteMCP:events_list
remoteMCP:helm_install
remoteMCP:helm_list
remoteMCP:helm_uninstall
remoteMCP:namespaces_list
remoteMCP:pods_delete
remoteMCP:pods_exec
remoteMCP:pods_get
remoteMCP:pods_list
remoteMCP:pods_list_in_namespace
remoteMCP:pods_log
remoteMCP:pods_run
remoteMCP:pods_top
remoteMCP:resources_create_or_update
remoteMCP:resources_delete
remoteMCP:resources_get
remoteMCP:resources_list
VertexAI Function Naming Requirements
According to the error message, function names must:
✅ Start with a letter or underscore
✅ Be alphanumeric (a-z, A-Z, 0-9)
✅ Allow underscores (_)
✅ Allow dots (.)
✅ Allow dashes (-)
❌ NO colons (:) - this is the issue
✅ Maximum 64 characters
Setup 1:
I have Claude desktop setup in windows and using the mcp-proxy.exe tool (installed on windows) able to call this MCP which is running in docker in another machine. Here is the sample config. And this works fine and is using the
Claude Sonnet 4modelSetup 2:
I have liteLLM and anythingLLM running as docker container and using the
Gemini-2.0-flashmodel. They are also configured with this MCP server. In both the scenarios, the functions / tools hosted by this MCP server is visible in both liteLLM and anythingLLM. However, when chatting both complain about the schema problem. So looks like this MCP server does not confirm to the stricter naming convention of Gemini models or more specifically VertexAI?The config in anytingLLM
This is what I found out from logs:
Logs from anythingLLM
❌ Non-Compliant Function Names
These function names violate VertexAI's naming rules because they contain colons (:), which are not allowed:
remoteMCP:configuration_view
remoteMCP:events_list
remoteMCP:helm_install
remoteMCP:helm_list
remoteMCP:helm_uninstall
remoteMCP:namespaces_list
remoteMCP:pods_delete
remoteMCP:pods_exec
remoteMCP:pods_get
remoteMCP:pods_list
remoteMCP:pods_list_in_namespace
remoteMCP:pods_log
remoteMCP:pods_run
remoteMCP:pods_top
remoteMCP:resources_create_or_update
remoteMCP:resources_delete
remoteMCP:resources_get
remoteMCP:resources_list
VertexAI Function Naming Requirements
According to the error message, function names must:
✅ Start with a letter or underscore
✅ Be alphanumeric (a-z, A-Z, 0-9)
✅ Allow underscores (_)
✅ Allow dots (.)
✅ Allow dashes (-)
❌ NO colons (:) - this is the issue
✅ Maximum 64 characters