One MCP API for any AI agent you already use
Mastheads runs an MCP server at https://api.mastheads.app/v1/mcp that gives an AI agent access to your newsroom. Connect ChatGPT, Claude, Cursor or Codex to it and they can read your publications, your articles, and the sources each article was written from. With a key you created for it, they can also start an article and publish one.
Connect ChatGPT, Claude, Cursor or Codex
get_article_sources: the tool no other newsroom's MCP answers
get_article_sources returns the pages an article was written from, the verdicts the desks recorded against it, and the editor of record. An agent asking it does not have to take the article on trust, and neither do you.
list_domainslist_articlesget_articleget_article_sourcesread_jobsread_quotaread_settingsread_subscriptionpropose_generateget_research_reportget_audit
write_articlewrite_articlesrun_researchrun_auditpublish_article
A key created with Read only, the default, can never spend or publish, and that permission is fixed when the key is made. The server is available on the Growth plan and up, with 120 reads a minute per key.
Questions about the Mastheads MCP server
What is the Mastheads MCP server?
The Mastheads MCP server lets an AI client access your own newsroom through a set of tools. The client reads an article directly from your account, along with the sources it was written from and the checks it passed, so you do not need to copy the article into a chat window.
Which clients can connect to the Mastheads MCP server?
Any MCP client can connect to the Mastheads MCP server. The server uses streamable HTTP and is stateless, with one POST per message. Its open protocol lets even a client we have never heard of connect. We document ChatGPT, Claude Code and claude.ai, Cursor, VS Code and Codex, each with the config block it needs. Every client gets the same set of tools.
How does the Mastheads MCP server authenticate?
The Mastheads MCP server has two authentication methods. Claude Desktop, claude.ai and ChatGPT sign in with your Mastheads login through OAuth, with a consent screen on your own dashboard, so no key is pasted anywhere. Every other client sends the same API key the pull API takes, as a bearer token. There is no second account and no separate password.
Can an AI client connected to Mastheads spend my articles?
An AI client connected to Mastheads can spend your articles only with your permission. Mastheads lists the write tools only to a key created with Full access, and the OAuth consent screen has one tick box, Let this app spend, which is off unless you turn it on. A read-only key sees the eleven reads and nothing else.
What can an AI agent read through the Mastheads MCP server?
An AI agent can use the Mastheads MCP server to read your publications, their articles, one article in full, and its provenance: the pages it was written from, the verdicts the desks recorded against it, and the editor of record. It can also read what is running now, what quota is left, how a domain is configured, and any finished research report or domain audit.
Which tools does the Mastheads MCP API have?
The Mastheads MCP server has sixteen tools: eleven reads and five writes. The reads, available to every key, are list_domains, list_articles, get_article, get_article_sources, read_jobs, read_quota, read_settings, read_subscription, propose_generate, get_research_report and get_audit. The writes are write_article, write_articles, run_research, run_audit and publish_article, and they are listed only to a key created with Full access.
How does an AI agent check where a Mastheads article came from?
An AI agent checks where a Mastheads article came from by calling get_article_sources through the Mastheads MCP server. The tool returns the pages the article was written from, the verdicts the desks recorded against it, and the editor of record, so the agent does not have to take the article on trust.
Can an AI agent publish an article to my site through Mastheads?
Yes, but only with a key created with Full access. The Mastheads publish_article tool puts a finished article live on the domain's connected site. A key created with Read only, which is the default, can never spend or publish, and that permission is fixed when the key is made.
How do I connect ChatGPT to Mastheads?
ChatGPT connects to Mastheads as a connector. Add the address https://api.mastheads.app/v1/mcp and sign in with your Mastheads login through OAuth 2.1, so there is no key to paste. You approve it on a consent screen at dash.mastheads.app, and Disconnect there ends it.
How do I connect Claude Code, Cursor, VS Code or Codex to Mastheads?
Claude Code, Cursor and VS Code use a config block with "type": "http", the Mastheads address https://api.mastheads.app/v1/mcp and your API key as a bearer token, which you create under Settings then Developer. Codex connects with one command, codex mcp add mastheads --url https://api.mastheads.app/v1/mcp, and stores the token itself. Claude Desktop and claude.ai sign in with your Mastheads login instead, with no key to paste.
How many apps and keys can I connect to Mastheads?
You can connect up to 10 apps to Mastheads through OAuth, counted separately from the 3 API keys you can create by hand. Connected apps appear under Settings then Developer, each with a Disconnect control that revokes access just as revoking a key does.
Is there a rate limit on the Mastheads MCP server?
Yes. The Mastheads MCP server allows 120 reads a minute per key. A key created with Full access can also start at most 20 paid actions an hour, where an article, a batch row, a report or an audit each count as one, and a batch can hold up to 100 rows.
What does the Mastheads MCP server cost?
The Mastheads MCP server costs nothing on top of your plan. It uses the same account, so an article started through it comes out of the same allowance as one started in the dashboard. The server is available on the Growth plan and up, with 120 reads a minute per key.