JoyfulWords now supports MCP. It provides a stable HTTP MCP Server endpoint for AI agents such as Claude Code and Codex: https://api.joyword.link/mcp. In practice, you can send that endpoint to your AI agent and ask it to install the service for you.
This article explains what that means in a real writing workflow: why an MCP server for AI agents matters, how Claude Code MCP setup fits into JoyfulWords, and what to check if the connection does not work.
So why would an article editor connect to an AI agent in the first place? If an AI agent could generate a perfect article on its own, the editor itself would no longer be necessary. But the reality is different. Many things cannot be fully explained in a single prompt. Sometimes the problem is not that AI is not smart enough. It is that we only discover the details of what we really want to say while we are writing.
In the past, writing an article meant reading references, adjusting wording, weighing sentences again and again, and spending a lot of time on the whole process. AI should help us shorten that time. It should not completely replace us.
That is why the more common workflow now looks like this: I first give AI my ideas, judgments, and the general direction I want to express, then let it help me build the skeleton of the article. At this stage, I do not expect AI to write something that truly sounds like "me" in one pass. What it does best is not finishing my expression for me. It is turning scattered thoughts into structure: what should come first, what should come later, which points need to be expanded, and where the article should hold back.
This matters a lot to me. Most of the time, the hardest part of writing is not that I do not know how to write. It is that I do not know where to begin. AI can quickly give me a first draft structure, turning the experience from "staring at a blank page" into "working with something I can revise." That step alone saves a lot of time.
What Is an MCP Server for AI Agents?
MCP, or Model Context Protocol, is a standard way for an AI agent to connect to external tools and data sources. Instead of pasting information back and forth, the agent can discover available capabilities through a server.
For JoyfulWords, the MCP server is not just a generic integration endpoint. It is a bridge between AI agents and a content workspace:
- articles and drafts
- material collection
- article improvement workflows
- image and content production context
- future publishing and repurposing workflows
That means the AI agent can support the work around the article, not only generate paragraphs in isolation.
Why HTTP MCP Matters
JoyfulWords exposes a standard HTTP MCP endpoint. This matters because many agent clients can connect to a remote HTTP service without requiring a local plugin build or a custom desktop bridge.
For Claude Code, the installation shape is:
claude mcp add --transport http --client-id joyfulwords-mcp-server joyfulwords https://api.joyword.link/mcp
After setup, restart Claude Code and run /mcp to check whether the JoyfulWords service is available.
AI Builds the Skeleton. I Make the Judgment.
AI-written articles often have one problem: they are too complete, and too flat.
Many sentences are not technically wrong. The structure may also make sense. But the writing still feels a little stiff. It is as if the information has been arranged neatly, but has not passed through a person's lived experience, voice, habits, or sense of what to keep and what to cut. This is especially obvious in sections that need personal color. If everything is handed over to AI, the article can easily become a pile of polished words.
So I wanted a place where I could save the articles AI writes, then edit them easily. JoyfulWords is the answer I built for myself. There are already many editors and note-taking apps on the market, but they do not feel smooth enough for this workflow, because they were born before AI became part of everyday writing. Many AI features are not built in natively.
JoyfulWords Helps the Article Grow Flesh
At this stage, JoyfulWords helps me most by moving an article from "it has a structure" to "it feels like my own article."
After AI gives me the skeleton, I open it in JoyfulWords and read through it again. Which paragraphs feel too templated? Which sentences are too hard? Which ideas need to sound closer to everyday language? I can edit them directly, or ask AI to revise a specific passage for me and try different tones: more direct, more restrained, more conversational, or better suited for social media.
I also use material search to add details to the article. A concrete example, an image, or a real scenario is often more persuasive than a whole paragraph of abstract explanation. Material search lets me find references on the web more quickly and bring them into the article, so the article is not just a stack of opinions.
If web images are not enough, or if they come with copyright issues or watermarks, which is now extremely common on Google, I can use one-click image generation to create new visuals with AI, or stylize existing images.
I can even use AI directly to generate the kind of infographic that is popular online right now. These images combine text, layout, and decoration into one visual, and they really can say more than a long block of words.
Common MCP Setup Issues
If an AI agent cannot connect to JoyfulWords MCP, check these points first:
- Wrong endpoint: the production MCP endpoint is
https://api.joyword.link/mcp. - Wrong transport: use HTTP transport when the client asks for a transport type.
- Client id mismatch: use
joyfulwords-mcp-serverwhen the client supports an explicit client id. - OAuth not completed: some clients need a browser authorization step before tools appear.
- Client not restarted: Claude Code may need to restart before the new MCP server appears in
/mcp.
These checks usually tell you whether the issue is installation, OAuth, or tool discovery.
What Humans Really Need Is Collaboration, Not Replacement
I do not really think of AI writing as "machines producing ten thousand articles a day." That kind of low-quality content is meaningless, and it will eventually be eliminated by platforms. The core of any platform is user retention, and users do not go online because they want to watch AI lecture them. If they wanted that, they could just ask AI directly.
So I think the best use case for mass-producing articles with AI is probably mechanical, fast, emotionless press releases.
The approach I believe in more is collaboration: AI helps me build the structure, and I continue polishing the language, adding materials, generating images, and making the final judgment myself. Which parts should stay? Which parts should be deleted? Which expressions need to sound more like me? Where do I need to add something that belongs to my own perspective?
The benefit of this partnership is that it does not sacrifice personal expression, and it does not throw away efficiency. I do not need to spend huge amounts of time organizing outlines, searching for materials again and again, or manually adjusting every sentence. But the final article can still keep my point of view, my rhythm, and my choices.
What Comes Next for JoyfulWords
I do not plan to add bulk AI article generation to JoyfulWords. The focus is managing and improving articles after an AI agent has generated them. So I will continue adding more AI features that make this workflow easier to use. If you notice any problems while using it, you can leave me feedback directly.
For JoyfulWords, I did not design a simple subscription model. Instead, it records the underlying usage cost directly, and every expense is traceable. I think one subscription is enough, and that subscription should be the AI agent. Everything else does not need to be another recurring bill.
That is it.
Best Regards~