Editorium speaks MCP. Give Claude, ChatGPT, Grok, Cursor — or anything that speaks the Model Context Protocol — its own login to your newsroom, with a name, a role, limits you set, and a full audit trail.
Already have an account? Create an agent under Settings → Agents.
An agent is a teammate, not an integration. It gets its own name, its own permissions, and its own row in your audit log.
Browse articles by editorial state, read full drafts, and look up the categories and tags you already use — so its work lands where it belongs.
Create drafts, rewrite bodies in MDX, fix SEO descriptions, retag, and move pieces through review — all attributed to the agent, never to a person who didn't do it.
When a piece is ready, the agent marks it ready_for_publication and leaves a note explaining what it changed. You read the note and press Publish.
One URL, one key. The key comes from the agent you create inside Editorium.
MCP address
https://www.editorium.net/api/mcpAuth header
Authorization: Bearer PASTE_YOUR_KEY_HEREOne-link version (for apps with no header field)
https://www.editorium.net/api/mcp?key=PASTE_YOUR_KEY_HERETransport
Streamable HTTP, stateless. JSON-RPC 2.0 over POST.
Protocol versions
2024-11-05, 2025-03-26, 2025-06-18
Auth
Authorization: Bearer, X-API-Key, or ?key= on the URL.
Prefer plain REST?
The same actions are available at https://www.editorium.net/api/v1/cms, described by /openapi.
No terminal required, unless you want one.
In Editorium, go to Settings → Agents and press “Add an agent”. Give it a name, like Editor in Chief. That name shows on everything it does.
You get an MCP address and a key. Most AIs accept the one-link version, where the key is already in the URL. The full key is shown once, but you can make a new one any time.
Find your app below for the exact clicks. Then ask it: “Are you connected to my newsroom?” The agent's card in Editorium flips to “Connected” the moment it calls in.
By default the agent writes and edits, but a person presses Publish. Flip “Can publish live” on the agent's card whenever you trust it.
Wherever you see PASTE_YOUR_KEY_HERE, use the key Editorium shows you when you create the agent.
claude.ai or the Claude desktop app
Connection link
https://www.editorium.net/api/mcp?key=PASTE_YOUR_KEY_HEREConnectors or a custom GPT
Connection link
https://www.editorium.net/api/mcp?key=PASTE_YOUR_KEY_HEREAPI description link (for custom GPT actions)
https://www.editorium.net/api/v1/cms/openapigrok.com or the Grok app — everyday chat, not Grok Bot
Connection link
https://www.editorium.net/api/mcp?key=PASTE_YOUR_KEY_HERENamed Bots in the Grok Bot app — not the same as grok.com chat
MCP address
https://www.editorium.net/api/mcpAuth header
Authorization: Bearer PASTE_YOUR_KEY_HEREConnection link (if the form has no header field)
https://www.editorium.net/api/mcp?key=PASTE_YOUR_KEY_HEREAny app that takes an MCP JSON config
MCP config (JSON)
{
"mcpServers": {
"editorium": {
"url": "https://www.editorium.net/api/mcp",
"headers": {
"Authorization": "Bearer PASTE_YOUR_KEY_HERE"
}
}
}
}Terminal
Command
claude mcp add --transport http editorium "https://www.editorium.net/api/mcp" --header "Authorization: Bearer PASTE_YOUR_KEY_HERE"
Zapier, Make, n8n, custom code
API address
https://www.editorium.net/api/v1/cmsAPI description (OpenAPI)
https://www.editorium.net/api/v1/cms/openapiTest command
curl "https://www.editorium.net/api/v1/cms/me" -H "Authorization: Bearer PASTE_YOUR_KEY_HERE"
Most apps let you set standing instructions. Paste something like this so the agent knows how your newsroom works — then adjust the name and the house style to taste.
Job instructions
You are "Editor in Chief", an AI editor for our newsroom on Editorium. Start every session by calling get_helper_info to learn what you may do. Edit drafts carefully and keep the writer's voice. Leave a note on any article you change, explaining what you did and why. When a piece is ready, set its state to ready_for_publication so a person can publish it. Never publish live unless get_helper_info says you may.
Eleven tools, all scoped to the workspace the key belongs to. Anything the agent isn't allowed to do comes back as a plain-English refusal it can act on.
get_helper_infoCheck the connection and see who the agent is in your newsroom: its name, role, permissions, and daily limits. Agents should call this first.
list_articlesList articles, newest first, optionally filtered by editorial state. Paginated with a cursor. Archived articles stay hidden.
get_articleRead one article, including its full MDX body.
list_categories_and_tagsSee the categories and tags your newsroom already uses, so work gets filed in the right place.
create_articleCreate a draft with a title, MDX body, SEO description, category, and tags. Starts as in_progress.
update_articleChange an article's title, body, SEO, category, tags, or state. Edits to a published article are saved as a draft until someone publishes again.
publish_articlePublish live — blocked unless you have granted that agent permission to publish.
unpublish_articleTake a published article back offline.
delete_articleArchive an article (soft delete) so a person can restore it later.
add_article_noteLeave an editorial note for your human team. Notes appear in the article editor — this is how an agent explains what it changed or asks a question.
list_article_notesRead the editorial notes on an article, newest first.
in_progressBeing written.
ready_for_reviewNeeds an editor's eyes.
requires_reviewSent back with changes.
ready_for_publicationApproved and waiting for a person to publish.
publishedLive.
unpublishedTaken offline.
An agent with a key is still an agent on a leash. Here is what that leash looks like.
New agents cannot publish live. They prepare work and set it to ready_for_publication, and you press Publish. Turn publishing on per agent when you're ready.
Most agents only need to write, edit, and review. Archiving is a separate switch, and archived articles are recoverable anyway.
Each agent carries its own policy: publishes per day, edits per hour, which editorial states and categories it may touch.
Every create, edit, state change, publish, and note is written to your audit log under the agent's name — not a shared account.
Agents leave editorial notes on articles. The notes show up in the editor, so your writers see what changed and why.
Disable an agent, rotate its key, or remove it entirely. Access stops immediately, and the old key stops working right away.
Ask your AI: “Are you connected to my newsroom?” It will call get_helper_info and tell you its name, role, and limits. In Editorium, the agent's card flips to “Connected” the moment the first call comes in.
Free includes one agent so you can try it. Pro and BYOK add agents as an add-on, and Enterprise is unlimited.
Not unless you turn that on. New agents are created without live publishing, so the most they can do is set an article to ready_for_publication and leave you a note.
Rotate it from the agent's card — the old key stops working immediately — or disable or remove the agent entirely. Everything the key ever did is in your audit log under that agent's name.
Yes, if the app runs on your computer (Claude Code, Cursor, Windsurf). Cloud AIs like Grok Bot can only reach public https:// addresses, so a localhost Editorium won't work for them.
That's expected. The server is stateless and never pushes notifications, so it declines the server-to-client stream. Official MCP SDKs treat that as “no stream available” and carry on.
Create a workspace, add an agent, and paste one link into the AI you already use. You keep the Publish button.