MCP vs skill?

[waldyrious]

Hi there! I have recently started to experiment with OpenClaw, and since there was no existing skill for it to connect to Workflowy, I created one, based on your CLI (it's just a SKILL.md file explaining how to use your CLI).

It's been working great so far, but I wonder if I'm reinventing the wheel or using it in a suboptimal way, considering that you already have an MCP set up. Would you recommend a different way of hooking your tool to OpenClaw? Or perhaps some tweaks to the SKILL.md file I created?

[mholzen]

Hi Waldir, thanks for reaching out!

Nice job with the skills file, thanks for doing that.

Comparing a skills file with an MCP server:

• from a performance stand point: I don't think there is much of a difference. With the skill, the agent will issue a new workflowy command every time, compared to sending a JSON payload representing the tool invocation to an already running process. Most of the time will be spend executing the command and making API calls to WorkFlowy so I don't think either models matters.
• from a token's consumption perspective: the MCP server defaults to producing JSON whereas the CLI defaults to producing text, (even though you can ask the CLI to produce JSON). The JSON output format most likely consumes more tokens although I haven't measured it. On the other hand, the JSON output has a higher fidelity to the content stored in WorkFlowy (because there is no transcription to Markdown happening), and contains more information (eg. each node is returned with created date, modified date as well as the layout info (see https://beta.workflowy.com/api-reference/#nodes-object )

The one aspect that could make a difference is the possible divergence between the SKILL.md file you created and the options available in the CLI. Your SKILL.md file would have to be kept in sync with new releases of the CLI and MCP server ( see https://github.com/mholzen/workflowy/releases) . One advantage of the MCP server is that it advertises its tools with description if all parameters at startup. So, they are always in sync. This divergence could produce some frustrating and difficult to troubleshoot issues.

But, I like being able to offer both approach and let the user choose. So, if you don't mind, I could incorporate your skills file into this repo, and ensure it is always in sync with the CLI, including updating https://clawhub.ai/ on every release. I already have a list of release tasks (https://github.com/mholzen/workflowy/blob/main/docs/RELEASE.md)

What do you think?

[waldyrious]

Thanks for the thoughtful and detailed response!

I think it makes sense to keep the skill file in this repo, and if you don't mind updating the ClawHub copy on every release, I'd be happy to transfer that skill's ownership to you as soon as that functionality is released on GitHub and/or NPM.

In the meantime, in case you haven't yet signed up at ClawHub, please do so to ensure the transfer will be possible once the functionality is in place.