Connect Aveiro to Any AI Agent in 10 Seconds — One MCP URL
Paste one URL into ChatGPT, Claude, Cursor, or VS Code and your agent can list your sites, edit pages, publish, draft newsletters, and prepare social posts — with OAuth sign-in and human guardrails built in.
Aveiro's hosted MCP server lives at one URL: https://www.aveiro.app/api/mcp/mcp. Paste it into any MCP-capable client — ChatGPT, Claude, Cursor, VS Code — sign in, pick your organization, and the agent can read and edit your sites, publish pages, and draft campaigns. No repo checkout, no local server, no API glue code.
The whole setup is a URL
Most integrations start with a checklist: install a CLI, clone a repo, generate keys, wire a webhook. Connecting an AI agent to Aveiro starts and ends with one line:That is Aveiro's hosted MCP server — the Model Context Protocol endpoint that exposes your sites, pages, newsletter, and social drafts as tools any MCP client can call. Paste it into your client of choice, authorize, and you are connected. Ten seconds is not a metaphor: copy, paste, sign in, done.There is nothing to host and nothing to update. The server runs on Aveiro's side and always matches the live API.
Connect from ChatGPT
ChatGPT supports remote MCP servers as connectors:
Open ChatGPT settings and go to Connectors.
Add a custom connector and paste https://www.aveiro.app/api/mcp/mcp.
Sign in with your Aveiro account and choose the organization to grant access to.
From that point, you can ask ChatGPT things like "list my Aveiro sites", "rewrite the hero copy on my homepage draft", or "publish the pricing page" — and it happens on your real, hosted site, not in a copy-paste buffer.
claude mcp add --transport http aveiro https://www.aveiro.app/api/mcp/mcp
Cursor
Add to Cursor from the dashboard's Connect agents card, or add the JSON config below
VS Code
Add to VS Code from the same card, or register the URL as an HTTP MCP server
https://www.aveiro.app/api/mcp/mcp
For clients that read a JSON config:
The dashboard's API tokens page has a Connect agents card with one-click install buttons for Cursor and VS Code and copy-ready snippets for everything else.
How sign-in works
OAuth-capable clients (ChatGPT, Claude) walk you through it: the first connection opens an Aveiro sign-in, you pick which organization the agent may act on, and the grant is scoped to that choice. No keys to copy.
Clients that authenticate with a header instead — scripts, CI agents, self-hosted runners — use a personal access token. Create one under Organization → API tokens with the scopes you want (for example publish:write) and send it as an Authorization: Bearer header.
Either way, the agent only gets the scopes you granted. Access is revocable from the dashboard at any time.
What can an agent actually do?
Everything runs through the same permission-checked API the dashboard uses. The tool set covers all three sides of connected publishing:
Area
Tools
What agents do with them
Sites & pages
list, create, edit, reorder, delete, publish
Read your file tree, draft MDX pages, restructure collections, push a site live
Newsletter
list campaigns, read analytics, create and revise drafts
Draft campaigns and pull open/click metrics — sending stays human-only
Social
draft posts, generate and attach media, per-platform captions
Prepare Instagram, Threads, and X posts for your review queue
Some concrete workflows people run today:
Edit your site from a chat. Ask ChatGPT to tighten your landing-page copy; the agent reads the draft with aveiro_get_page, updates it with aveiro_update_page, and you review the diff in the editor before publishing.
Ship a page from your editor. In Cursor or Claude Code, ask the agent to create a changelog entry — it writes the MDX, files it into the right collection, and calls aveiro_publish_site when you say go.
Draft the launch email. The agent creates a campaign draft with subject, preview text, and body; you open Audience → Campaigns, tweak, and hit Send yourself.
Prepare the social push. The agent drafts platform-specific captions and generates media, then submits the post to your review queue.
Audit what changed. Ask for the recent change history on a site before you publish — useful when several agents (or teammates) touch the same workspace.
Guardrails: agents draft, humans decide
Wiring an AI agent to your website and email list is only a good idea if the blast radius is controlled. Aveiro's guardrails are structural, not honor-system:
Scoped access. Tokens and OAuth grants carry explicit scopes (publish:write, newsletter:write, social:write, …). No scope, no tool.
No autonomous email. There is no API route and no MCP tool that sends or schedules a campaign. Agents draft; a human clicks Send in the dashboard.
Social posts go through review. Drafts land in an approval queue; a human approves and schedules publication.
Drafts before live. Page edits land in your draft tree. Publishing is a separate, explicit step — and one you can keep for yourself by withholding publish:write.
So "access and edit sites from ChatGPT" means exactly that — with the same review points you would demand from a human collaborator.
FAQ
Do I need to run anything locally?
No. The hosted server is the default path. A stdio server exists in the open-source repo for local development, but the URL is all most people ever need.
Which clients are supported?
Any MCP client that speaks Streamable HTTP. ChatGPT, Claude.ai, Claude Desktop, Claude Code, Cursor, and VS Code are the ones we test against — including the client-specific quirks we normalize server-side.
Can an agent delete my site?
Destructive operations require their own scopes, and every tool declares read-only/destructive hints so well-behaved clients ask before acting. Grant narrowly and expand later.
Is there a plain REST API too?
Yes — the MCP tools are a thin wrapper over /api/v1, so anything an agent can do, your own scripts can do with the same token.
