One MCP connector, any AI client.
Pagepost is a standard remote MCP server. If your assistant can add a remote MCP server and sign in through OAuth, it can publish an HTML dashboard or a Markdown report as a real URL — and load that page back later to revise it.
There is one endpoint for every client. Nothing here is vendor-specific.
Connection details
- Endpoint
https://pagepost.app/api/mcp- Transport
- Streamable HTTP (JSON-RPC 2.0 over POST)
- Protocol revision
2025-06-18- Auth
- OAuth 2.1 — dynamic client registration, authorization code + PKCE
- Scope
publish- Discovery
/.well-known/oauth-protected-resourceand/.well-known/oauth-authorization-server- Tools
- publish_file, update_file, search, fetch
- 1
Add a remote MCP server
In your client's MCP or connector settings, add a remote server with the streamable-HTTP transport and the URLhttps://pagepost.app/api/mcp. If the field expects a base URL rather than a full endpoint,https://pagepost.appresolves to the same place. - 2
Sign in once
The client opens a browser window for OAuth — sign in with Google or Slack. There is no key to generate or paste. The token is cached, so this is a one-time step per client. - 3
Ask it to publish
Have the model write an HTML dashboard or a Markdown report, then say “publish it to Pagepost”. It calls the connector and hands back a live URL you can paste anywhere.
What the connector exposes
publish_file
Takes an HTML or Markdown body and returns a live URL. The one tool most sessions need.
update_file
Rewrites a page you already published, in place. The URL you shared stays the same and everyone holding it sees the new version.
search
Lists the pages you have published, matched on title or filename, so the model can find the one you mean without you pasting an ID.
fetch
Returns a page's real source. This is what makes load-then-edit work: fetch, revise, update_file, same URL.
Before you start
Two requirements, and they rule out a real slice of clients: the client has to support remote MCP servers (not just local ones launched over stdio), and it has to run the OAuth flow — Pagepost has no static API key to paste. Everything else, including which model you point at it, is up to you. If your client only speaks stdio or static headers, upload the file directly and skip the connector.
Step-by-step for a specific client
Same endpoint, different settings screens. These pages show exactly where the URL goes:
Other clients that speak remote MCP
These have no page of their own yet, but they support remote MCP servers and can complete an OAuth sign-in, which is all Pagepost needs. The wording differs — some call it a connector, some an MCP server, some just paste a URL into a config file:
- Cursor — MCP settings, added as a remote server URL
- VS Code with GitHub Copilot — MCP servers in agent mode, configured per workspace or globally
- Windsurf — Cascade's plugin and MCP configuration
- Cline — Remote MCP servers, added from the MCP panel
- Continue — MCP blocks in your assistant config
- Zed — Context servers in settings
- Goose — Extensions, added as a remote extension
- LibreChat — MCP servers in librechat.yaml
- Cherry Studio — MCP servers in settings
Checked August 2026. This space moves fast — if your client isn't here, look for “add remote MCP server” or “custom connector” in its settings and check that it can sign in through OAuth rather than only a pasted API key.
Publish, then keep editing
The connector is not write-once. fetch hands the model the real source of a page you published earlier, and update_file writes the revision back to the same URL — no new link to re-send.
Scoped to your account
The OAuth token identifies you, and every call runs as your account. A model can only search and fetch pages you own.
One library, every source
Pages published from an assistant sit in the same dashboard as your web uploads and anything the Slack bot posted. Rename, revoke, or expire them from one place.
Questions
Which MCP clients work with Pagepost?+
Any client that supports remote MCP servers over streamable HTTP and can run an OAuth flow. We document step-by-step setup for Claude, ChatGPT, Gemini and DeepSeek. Cursor, VS Code with GitHub Copilot, Windsurf, Cline, Continue, Zed, Goose, LibreChat and Cherry Studio also support remote MCP servers and use the same URL — there is nothing vendor-specific in the server.
Is there a different endpoint per AI client?+
No. There is one server at https://pagepost.app/api/mcp and every client connects to it. MCP is a standard, so the only thing that differs per client is where you paste the URL.
Can I connect with an API key instead of OAuth?+
No. Pagepost issues tokens through OAuth only — dynamic client registration, authorization code with PKCE, and refresh tokens — so there is no long-lived key to paste. A client that only accepts a static Authorization header cannot connect yet.
What can the connector do once it is connected?+
Four tools. publish_file turns an HTML or Markdown body into a live URL, update_file rewrites a page in place at the same URL, search lists the pages you have published, and fetch returns a page's source so the model can revise it.
Can a model see pages other people published?+
No. Every tool call runs as the account that authorized the connector, and search and fetch only ever return pages that account owns. A page belonging to anyone else answers the same way as one that does not exist.
Are there size limits?+
Publishing accepts documents up to 5 MB. fetch returns at most the first 400,000 bytes of a document and says so in the response, because a larger body would exhaust the model's context before you saw a result.
Do published pages stay editable outside the connector?+
Yes. Anything an assistant publishes appears in your Pagepost dashboard next to your manual uploads, where you can rename, revoke, expire, or edit it in the browser. The connector is one way in, not a separate silo.
Not using an assistant at all? Upload directly or use it from Slack.