MCP — the Model Context Protocol — lets an AI tool you already use work directly with your CamBuildr account. Claude, ChatGPT, Cursor, VS Code Copilot and Gemini CLI can look up your landing pages, mails, target audiences, workflows and contacts, create and change them, and hand content jobs back to Cora.
The short version: Cora works inside CamBuildr. MCP lets an outside AI tool work on CamBuildr. They are not either/or — the MCP connection can hand a content job back to Cora, so your AI tool can create a landing page and then ask Cora to write and build its content.
For what Cora does inside the product, see What Cora can do — the full feature guide.
Before you start
MCP is switched on per account. If you do not see MCP in your settings menu, write to us in the chat bubble and we will enable it.
Write access is a second switch. Without it, only the read tools and reference data are offered to your AI tool; the write tools appear greyed out as Unavailable. Ask us for it when you want your AI tool to create and change things, not just read them.
The AI content tools need AI switched on in Settings > AI — they hand work to Cora and spend AI credits. See Setting up AI in CamBuildr.
The connection acts as you. Whoever authorises the connection is the person the AI tool acts as, so treat it like handing someone your login.
Step 1: Copy your server URL
Go to Settings > MCP. The Connection card at the top shows your account's own MCP Server URL — your CamBuildr address with /mcp on the end. Copy it with the button next to it.
Every account has its own URL. It is the only thing you need for most clients: there is no token to paste in.
Step 2: Connect your AI tool
The same card has a tab for each client with the exact steps. Here they are for reference — replace YOUR-SERVER-URL with the URL you just copied.
Claude
Open Settings → Connectors and click Add custom connector.
Paste your server URL as the remote MCP server URL.
Leave the OAuth Client ID and Client Secret fields blank.
Save, then click Connect and log in when your browser opens.
Claude Code
Run this once in a terminal:
claude mcp add --transport http cambuildr YOUR-SERVER-URL --scope user
ChatGPT
Open Settings → Connectors and turn on developer mode.
Add a connector with your server URL.
Custom connectors need a ChatGPT Pro, Team, Enterprise or Edu plan.
Cursor
Add this to ~/.cursor/mcp.json:
{"mcpServers": {"cambuildr": {"url": "YOUR-SERVER-URL"}}}
VS Code / Copilot
Add this to .vscode/mcp.json — note the key is servers, not mcpServers:
{"servers": {"cambuildr": {"type": "http", "url": "YOUR-SERVER-URL"}}}
Gemini CLI
Add this to ~/.gemini/settings.json:
{"mcpServers": {"cambuildr": {"type": "http", "url": "YOUR-SERVER-URL"}}}
Windsurf, Cline and Continue
These cannot open a browser to sign in on their own. Point them at your server URL through the mcp-remote bridge — run npx mcp-remote followed by your server URL as their MCP command — instead of connecting to the URL directly.
Signing in
Every client authenticates the same way: the first time it uses a tool it opens a browser, you log into CamBuildr, and an authorisation screen shows you which tool is asking and what it wants access to. Approve it once and you are connected.
There is nothing to copy, and no key to keep safe.
Access tokens — only if your client cannot sign in
The Access tokens card exists for the handful of clients that cannot do the browser sign-in above. Most AI assistants never need one.
Click Create token and give it a name you will recognise later, such as Claude Desktop or VS Code.
Copy it immediately. It is shown once and never again.
The list shows when each token was created and when it was last used, so you can spot ones nobody uses.
Revoke takes access away straight away — the connected AI tool stops working the moment you do.
Treat a token like a password: it acts as the person who created it.
Choosing what your AI tool may use
The Enabled tools card decides what is offered over the connection. Everything is on by default, and one switch per entry is all it takes. Anything you switch off is simply never offered to the connected tool.
They are grouped so you can make the decision at the right level:
Read tools — look up data in this account. They never change anything.
Write tools — create and change records. They act on live entities straight away, without a preview step.
AI content tools — hand a piece of work to Cora, who then edits the content herself and spends AI credits.
Guided workflows — multi-step recipes your AI tool can start on request. They only sequence the tools above and can do nothing those tools cannot.
Reference data — read-only background about your account that a client can attach once instead of fetching it again every turn.
Entries marked Unavailable are ones your account is not set up for: write tools without MCP write access, AI tools without AI switched on. They are never offered to a client until that is fixed.
What CamBuildr exposes over MCP
Read tools
Read person — one person record by ID, with tags, addresses and a donation summary.
List people — optionally narrowed by a search term or a target audience.
List content — landing pages, campaign mails, automated mails and pages.
List target audiences — the non-archived ones, with their people count.
List workflows — with their state and whether they have a published graph.
List tags, List custom fields, List trigger actions — the building blocks other tools expect, with their IDs and allowed values.
Get landing page, Get campaign mail, Get automated mail, Get target audience, Get workflow — one record in full.
Search media library — the images and files this account already has.
Get hosted page contract — the authoring rules for AI-hosted pages.
Search and Fetch — one text query across everything, and one record back by ID. These exist for clients such as ChatGPT that drive a server through a search/fetch pair rather than per-type tools.
Write tools
Create and update landing page, campaign mail, automated mail, target audience and workflow — the record and its settings. Content is filled in separately, usually by Cora.
Set trigger action — bind an automated mail to the event that fires it.
Set workflow graph and Set workflow start condition — replace a workflow's nodes and edges and publish them as a new version, and replace the event that enrols people. See Workflows for what a graph is made of.
Add media library item — copy a file from a public URL into your media library so a page can reference it.
AI-hosted landing pages — create a page from raw HTML that CamBuildr hosts without the editor, add A/B variants to it, replace its HTML, and define a signup form on it. Useful when your AI tool wants to write the markup itself.
AI content tools
Instruct AI assistant — give Cora an instruction in plain language to write or rewrite the content of a landing page, mail, target audience or workflow. This is how an external tool gets real content built: it creates the shell, then asks Cora to fill it.
Guided workflows
Build a landing page — create the page shell, then have Cora fill its content.
Build an automated mail — create the mail, bind it to a real firing event, fill its content, and only then activate it.
Build a target audience — create the group, have its rules written, then check how many people it matches.
Build a workflow automation — create the workflow, set its start condition and graph, and activate it last.
Audit this account — a read-only walkthrough of every landing page, mail and target audience, flagging what looks unfinished, unbound or inactive.
Reference data
Account profile — who this connection belongs to: name, public base URL, content languages and which MCP features are in effect.
Tool catalogue — the exact names of every tool, prompt and resource your account exposes.
Trigger action catalogue — the internal events an automated mail can fire on, grouped by category.
Workflow building blocks — the node kinds, action shapes and real mail, tag and campaign IDs a workflow graph may reference.
Good to know before you let a tool write
Write tools act immediately. There is no preview and no approval step like Cora's Plan first mode. What the tool asks for is what happens.
There are no checkpoints here. Cora's Restore checkpoint is a feature of the chat inside CamBuildr, not of the MCP connection. Landing page and email history still works as usual.
Contact data is reachable. Read person and List people send real supporter records to whichever AI tool you connected. If that is not what you want, switch those two off in Enabled tools.
The AI content tools spend AI credits, because they run Cora. The read and write tools do not.
Things arrive as drafts. New landing pages are unpublished, new automated mails and workflows are inactive. Nothing goes live to your supporters without you publishing or activating it.
Troubleshooting
No MCP entry in the settings menu — MCP is not switched on for your account yet. Write to us in the chat.
Your AI tool only sees read tools — MCP write access is a separate switch. Ask us to enable it.
The AI content tools are missing or marked Unavailable — AI is not switched on. Enable it in Settings > AI first.
The browser never opens to sign in — your client probably cannot do it. Use the
mcp-remotebridge, or create an access token.VS Code does not pick the server up — check the key is
serversand that"type": "http"is present. Cursor is the one client that does not need the type.A tool your AI cannot find — check it is switched on under Enabled tools, then ask the tool to re-read the Tool catalogue.
Related articles
What Cora can do — the full feature guide — everything the built-in assistant does inside CamBuildr
Setting up AI in CamBuildr — enabling AI, API keys, customer instructions, allowed domains
Using Cora, the AI assistant in CamBuildr — basics and prompt writing
Workflows: automating what happens after someone acts — what the workflow tools above build
