When I added an MCP server to XreplyAI (a social media scheduling tool), I wasn't just adding a feature — I was betting that the people most likely to care about consistent social presence are the same people already living in Claude Code or Cursor.
So far that bet feels right.
Here's what the setup looks like:
{
"mcpServers": {
"xreplyai": {
"command": "npx",
"args": ["-y", "@xreplyai/mcp@latest"]
}
}
}
After that, Claude Code has tools to create drafts, schedule posts, and publish across 8 platforms — X, LinkedIn, Instagram, Threads, YouTube, Pinterest, Bluesky, TikTok — without leaving the editor.
What I learned building it
The MCP TypeScript SDK is well-designed and the protocol is cleaner than I expected for something this new. The trickier parts were:
1. Remote MCP over HTTP — I deployed to Cloudflare Workers with OAuth 2.1. The spec exists but examples are sparse. Durable Objects handle session state cleanly.
2. Guardrails matter — "publish a post" is a destructive tool. I added a draft-first flow so nothing goes live without an explicit publish call.
3. Voice is what makes it non-trivial — XreplyAI trains on your tweet archive, so when you ask Claude to "write a post about what I just shipped," it generates in your style, not a template. That's what separates it from just calling a REST API from a prompt.
The distribution angle
MCP tools feel like a new surface for early SaaS adoption. Low install friction, high-intent audience (developers deep in AI workflows), zero cold outreach needed. Worth thinking about if you're building a tool that developers would actually use.
Try it free: https://xreplyai.com?utm_source=devto&utm_medium=social&utm_campaign=feature-2026-06-01
Top comments (0)