maxclicks MCP Server
Work with your maxclicks workspace from an MCP-compatible assistant. Ask it to inspect an audience, update a customer record, or use a template your team has prepared. The assistant calls the API with the permissions you granted when you connected it, or those of the API key you gave it.
The server supports local stdio through the maxclicks-mcp npm package and remote Streamable HTTP at https://mcp.maxclicks.ai/mcp. Choose the transport your assistant supports and follow the setup instructions.
Current API coverage
The tools use the Public API for customer records, events, prepared campaigns, workflows, and webhooks. Follow each operation’s response and status fields: an accepted request can still require event readiness checks, deletion polling, or delivery tracking.
Connect your assistant
Add the server to your client
Add
https://mcp.maxclicks.ai/mcpas a remote MCP server in your assistant. In Claude Code, runclaude mcp add --transport http maxclicks https://mcp.maxclicks.ai/mcp, then/mcp. For a local stdio setup, use the local configuration.Approve the connection
Your browser opens maxclicks. Sign in, choose the workspace, and choose what the assistant may do: Read, Write, and Admin for suppressions if it asked for it. Check where the page says it Returns to, then click Allow access. An assistant that cannot sign in can use an API key from Settings, Developers, API keys instead, sent in an
Authorization: Bearerheader.Check it landed
Ask the assistant who it is connected as. It reports the workspace and the permissions it carries, which catches a connection to the wrong workspace before it writes anything.
What the assistant can do
It gets a tool for most of what the API can do: looking up and writing contacts and objects, firing events, building and counting segments, sending a template or a broadcast you drafted and reading back how it performed, driving workflows, and managing webhooks and suppressions.
Schemas, attributes, template and broadcast content, domains, senders and topics are read-only through the assistant. You author them in the app, and the assistant reads what you built.
Every tool acts as the person who connected the assistant, with the permissions they granted and their current role. Suppression tools require admin; other operations require the documented read or write permission. A missing permission returns 403. See authentication.
The MCP server reference is the full picture: every tool by name, the local stdio setup with a ready-made Claude Desktop config, the consent page, and how the server checks each request.
Sending is real. If you ask an assistant to send a broadcast, it sends to live people, spends credits, and cannot be recalled. Connect an assistant only to a workspace you are happy for it to act in, leave Write unchecked when it only needs to read, and be specific about what you want before you say send.
Going further
- MCP server reference: every tool, both run modes, the consent page, and how each request is checked.
- Build with AI: the other ways to point an agent at maxclicks, including the agent skills and the CLI.
- Authentication: how API keys are scoped and how an app connects with OAuth.
- Rate limits and errors: what the assistant runs into when it moves too fast or gets something wrong.
Questions or problems: [email protected].