Developers
Mentionwell MCP servers
Last updated
Ask the assistant you already work in for a blog post, and it publishes one to your site. Connect with one URL and a browser sign-in: nothing to install, no key to paste, and nothing to migrate.
Connect with one URL
Add Mentionwell as a custom connector and paste this address. Your browser opens, you sign in (or sign up) and click Allow. That’s the whole setup.
https://app.mentionwell.com/mcp
In Claude and Claude Desktop: Settings → Connectors → Add custom connector.
In ChatGPT: Settings → Connectors, add a custom connector. In Cursor, add it to
~/.cursor/mcp.json as {"mentionwell": {"url": "https://app.mentionwell.com/mcp"}}.
In Claude Code:
claude mcp add --transport http mentionwell https://app.mentionwell.com/mcp Each connection appears under Account → API tokens in the dashboard, where you can disconnect it at any time.
Or run it locally in two commands
You need an account, and this makes one: the login command opens your browser, signs you up if you are new, and stores a token on your machine. Nothing goes in your MCP config, and nothing gets copied out of a settings page.
npx mentionwell-cli login Then add the server. This is the whole config:
{
"mcpServers": {
"mentionwell": {
"command": "npx",
"args": ["-y", "mentionwell-mcp-account"]
}
}
}
That file is ~/.cursor/mcp.json in Cursor, your Claude Desktop config
in Claude, and ~/.codex/mcp.json in Codex. Restart the client, connect
the site you want to publish to, and ask for a post in plain language:
write a post on private label vs white label,
and ship it to landfallbrands.com The assistant researches, outlines, drafts, runs an editorial pass, generates the images and publishes — roughly twelve minutes — then hands back the live URL. It names your site by domain, slug, or the git remote of the repository you are sitting in, so from inside a project you rarely have to say which site you mean.
Running in CI, or prefer an explicit key? Set MENTIONWELL_TOKEN in the
server’s env and it takes precedence over the stored login.
What the assistant can do
| Tool | What it does |
|---|---|
create_site, get_setup_status | Add a new site from its domain: scan, brand profile and article plan. |
write_article | Research, write, illustrate and publish an article on a topic you name. |
get_article_status | Follow a run to its live URL. |
list_sites, get_site_status | See the sites on the account and how each one is connected. |
get_brand_context, update_brand_context | Read and adjust the audience, voice and CTA an article is written to. |
probe_site, start_connect, advance_connect, verify_webhook | Connect a new site and prove delivery works, without leaving the chat. |
import_posts | Pull an existing blog in from RSS, wp-json or Ghost. |
Publishing schedules, billing, quotas, the editorial calendar and the citation dashboards live in the app. The assistant is the fastest way to ask for a piece of work; the dashboard is where you run the programme.
The two servers
| Package | Auth | What it exposes |
|---|---|---|
mentionwell-mcp-account | Your stored login, or a PAT (mw_pat_*) | Everything in the table above: writing and publishing articles, sites, brand context, Connect, delivery checks, imports. |
mentionwell-mcp-docs | None — public | Answers Mentionwell questions from the official docs, so “wire Mentionwell into my Next.js project” produces working code instead of a guess. |
Why this matters
Publishing is the step that never got easier. Everything else in the stack moved into the editor you already work in, while putting an article on your own site still meant a dashboard, a CMS and a deploy. This removes that trip: the request happens where the thought happens, and the article lands on your site.
With mentionwell-mcp-docs installed too, “wire Mentionwell into
my Next.js project” produces working code on the first try, because the docs
are in the agent’s context rather than its training data.
Questions
What's MCP?
Model Context Protocol — an open standard introduced by Anthropic that lets AI agents connect to external tools and data via a uniform JSON-RPC contract. Claude Desktop, Cursor, ChatGPT (via experimental connector), and most agent frameworks support it natively.
Do I need an account to use mentionwell-mcp-docs?
No. The docs MCP is no-auth — install it and any MCP-aware client can ask it Mentionwell questions. It returns answers grounded in our official docs (no hallucinations from training-data drift).
What does mentionwell-mcp-account let me do?
List sites and site status, read and update a site's brand context, probe a domain, run the agentic Connect flow, verify webhooks, and import existing posts as drafts. PAT-authenticated (mw_pat_…). Designed for an AI agent driving Mentionwell onboarding and delivery on your behalf. It also writes: name a topic and a site and write_article runs the full pipeline and publishes the finished post, while get_article_status reports progress and hands back the live URL.
How do I connect Claude, ChatGPT or Cursor?
Add a custom connector with the URL https://app.mentionwell.com/mcp and sign in when your browser opens. Nothing to install and no key to copy. Prefer a local server? Run npx mentionwell-cli login and add mentionwell-mcp-account to your MCP config.