1. The problem: Allow the use of the existing codex/claude/antigravity cli subscription account to avoid paying seperate api cost
2. Simple solution: Use of existing ai subscription account inside SCM
thank you
Funny you mentioned this, as I have added claude code to our chat panels.
I will then add codex.
But these are chat only, will need to create a copy to run on normal AI operations for content gen as well.
2 Likes
That’s great, will test them… thank you so much !!!
A SCM skill with references and scripts (API endpoint wrapper) would be great, haven’t got the time to create this. So users can trigger inside their agent harness.
Yes the plan is to enable proper MCP support for SCM tools after the SEO Agent feature branch is released.
The reason why is once I give the agent access, the same code can be re-used as the MCP server.
Biggest roadblock is the pagepiple line. Although we do have a better working schema for it, it will require extra work so that LLM can work with it.
Previous attempts with ASK AI, and giving it commented JSON isn’t 100% full proof.
1 Like
IMHO mcp would be overkill because scm already have live API endpoint, agent only need instructions, that’s where skill come in.
A base skill can be customized for each project/website/domain, a human/agent can edit and iterate based on the project scope requirements.
So a skill utilization would be,
download scm skill → copy to project folder .agents/skills → edit/customize/iterate along the project lifetime. Rinse and repeat for new project.
MCP is one server for all project, customization in AGENTS.md not as flexible as skill and for SCM it would require 2x call first call mcp then call live API.
Not that I against MCP, I use it but not as much as skill. My mcp best use case is local tts speak tool. Load MCP once (keep the tts model alive and warm), and my agents can use the speak tool whether I use cc/codex/opencode or other mcp compatible clients.
Just my 5c 
True. A skills md will allow you to customise.
The api though needs some upgrades as it isn’t self documenting enough or self discoverable for it to be used purely with a skills doc easily. It can be done though with a lot of massaging and examples.
Eg: ai needs an endpoint to list tool calls with schemas.
Api doesn’t have that, only a hard api doc online somewhere, which can go stale easily and doesn’t provide a proper schema.
I think there is also basic plumbing missing in there, like streaming task progress or what not.
But when I say mcp I mean closer to tool registry + execution.
The AI goes directly to mcp to find and run tools, there won’t be another api layer for it.
But need some user case study for this, when we get closer to mcp stuff I’ll let you know and you can tell me what you want to do with skills.
1 Like
Does the postman collection/scm api auto generated when compiling scm build? I’m not knowledgeable about electron.
Does your version of MCP is built-in to scm so it’s separated from api?
My initial assumption is scm → api (source of truth) → mcp/skill as wrapper (maintenance) → agent.
1 Like
Postman is static doc, needs me to update when api changes.
The api is sot, but it’s not AI friendly.
I’m building agent inside scm now and what I discovered is you need a different set of interfaces that the current api doesn’t provide.
So while I build out the agent, most of what is there can be transferable to a mcp.
So what we will have is mcp, with tools and executions and it will sit separate to the api.
But this is still planning stage and I won’t know the exact details until I walk through it.
But what I can say is that the current api is lacking, especially schema stuff.
Page pipeline has too many params that aren’t self documented so the llm has issues using.
You are correct though that a skills md can be used to teach llm how to call api even with its current drawbacks.
What I propose is after agent build out I can take my learnings there and maybe we can build api/mcp such that the skills md will be telling ai what you want it to do and less of exactly how to do it.
Ie tell ai to create an article, without having to tell it specifically to call page pipeline and set articleKeywords=[“cat”]
1 Like
Sounds like a lot of work, a major release, but I bet it worth the time and experience.
I initially propose skill because building mcp is a lot of work, especially when you are solo. I treat skill with bells and whistles(docs → references folder, custom js → scripts folder) as prototype for later mcp development.
I also see the trend of skill, cli and mcp bundled that can be used separately or as orchestration/workflow.
For inspiration/reference, check out executor it’s on my todo/watch list for managing mcp servers, specific tools (enable/disable) and OpenAPIs.
Yes it is a bit of work to get it going.
Also means when agent update drops the homepage in its entirety will need updating as well.
See how it goes.
1 Like