Description
I've got an MCP server that i've written in golang, supports streamable http endpoint, endpoint is at /mcp. However I can't seem to figure out how to write the runtime block in agentcore.json that will allow it to function.
It's main issue seems to be:
- runtimes[0].entrypoint: Must be a Python (.py) or TypeScript (.ts/.js) file path with optional handler (e.g., "main.py:handler" or "index.ts")
Is there some way for me to 'bring my own docker container' to this process and just inject a docker image path for local or remote fetching? I understand that there might be parts of the tool that would be less functional but this tool seems pretty limited.
I've tried several variants on configs including some references to attributes like transport, port, and path which is great. The AI suggestions are pointing to old documentation for the previous version of this framework so I doubt they work anymore. However it seems like entrypoint must be present in the runtime config and must be either a .py or .ts/js reference which seems like an odd requirement.
My current runtime muddle looks loosely like this:
{
"name": "mymcp",
"build": "Container",
"buildContextPath": ".",
"entrypoint": "my-mcp",
"codeLocation": "cmd/",
"dockerfile": "./Dockerfile",
"transport": "http",
"port": 8080,
"path": "/mcp",
"networkMode": "PUBLIC",
"instrumentation": {
"enableOtel": false
},
"protocol": "MCP",
"stateless": true
}
if I lie to it and tell it it's entrypoint is my-mcp.py then I get a container build error with no details. Which is also one of the most user-hostile things I've encountered so far. Not even echoing the error message?
Acceptance Criteria
- Support for some kind of bring-your-own-container or entrypoint validation that does not require one of your 'blessed' languages.
- Better errors all around, If I have an error in my config 'can't find the project' is not acceptable, the output from
agentcore validate is.
Additional Context
No response
Description
I've got an MCP server that i've written in golang, supports streamable http endpoint, endpoint is at /mcp. However I can't seem to figure out how to write the runtime block in agentcore.json that will allow it to function.
It's main issue seems to be:
Is there some way for me to 'bring my own docker container' to this process and just inject a docker image path for local or remote fetching? I understand that there might be parts of the tool that would be less functional but this tool seems pretty limited.
I've tried several variants on configs including some references to attributes like transport, port, and path which is great. The AI suggestions are pointing to old documentation for the previous version of this framework so I doubt they work anymore. However it seems like entrypoint must be present in the runtime config and must be either a .py or .ts/js reference which seems like an odd requirement.
My current runtime muddle looks loosely like this:
if I lie to it and tell it it's entrypoint is my-mcp.py then I get a container build error with no details. Which is also one of the most user-hostile things I've encountered so far. Not even echoing the error message?
Acceptance Criteria
agentcore validateis.Additional Context
No response