We're live in beta. Earn up to 12,000 credits by signing up today. Get started

ServiceM8 Add-on Builder MCP
for AI agents

FloConnector is a hosted ServiceM8 Add-on Builder MCP server. Securely connect Claude, ChatGPT and other AI clients to ServiceM8 Add-on Builder. Ask in plain English; it runs the 11 tools for you, behind one endpoint, with you deciding what it can touch.

Beta. Connect it today and every tool below is live. It is new enough that we are still testing it against real ServiceM8 Add-on Builder accounts, so tell us if something looks off and we will fix it fast.

Built in-house. We wrote this connector ourselves and call ServiceM8 Add-on Builder directly, with no third party in the path. Nothing in a request or response is retained by anyone but ServiceM8 Add-on Builder.

Single connection. One ServiceM8 Add-on Builder organisation per workspace. Connect another in a separate workspace.

Tools

Everything your AI can do in ServiceM8 Add-on Builder

11 tools your agent gets the moment ServiceM8 Add-on Builder is connected.

11 tools

  • guide read

    Add-on builder guide

    How to build a ServiceM8 add-on. Call this before writing or changing any add-on: with no arguments for the overview (start there), with `topic` for one subject (manifest, events, responses, triggers, ...), or with `resource` for one ServiceM8 API resource's endpoints and fields.

  • list_addons read

    List add-ons

    List every add-on in the connected ServiceM8 account and which ones FloConnector built. Only those can be read in full or changed here; the rest (often other vendors' add-ons) are listed so you know they exist, and must be left alone.

  • get_addon read

    Get an add-on

    Get one add-on's ServiceM8 status and, if FloConnector built it, its saved draft and what is live now (manifest and handler code). ServiceM8 itself never returns an add-on's code. Read this before changing an add-on.

  • get_logs read

    Get add-on logs

    Collect a FloConnector add-on's live log output for a few seconds (default 30, max 45) and return it, including runs from the last few seconds before the call. Ask the user to trigger the add-on while this runs (click its button, or change a field it watches; webhook runs arrive about 17 seconds after the change), and call it again to keep listening. 'No logs yet' means it has not run since it was deployed.

  • list_secrets read

    List add-on secrets

    List the NAMES of a FloConnector add-on's secrets (values are never returned). The add-on reads each as process.env.NAME.

  • save_draft write

    Save an add-on draft

    Check and save an add-on's manifest and handler code as a DRAFT. Nothing reaches ServiceM8 and nothing goes live. Read servicem8addons_guide before writing an add-on. Checks: ServiceM8's manifest schema, a slug starting 'flc-', scopes within FloConnector's allowed list, buttons and menu entries that use 'event', and code that compiles as one CommonJS file with no npm packages (compiled only, never run). Saving again replaces the draft; what is live stays live until servicem8addons_deploy_addon. Show the user the review this returns before deploying.

  • deploy_addon write

    Deploy an add-on

    Deploy a FloConnector add-on's SAVED draft to ServiceM8. It goes LIVE in the account straight away: its buttons appear and its webhooks start firing. Only call this after the user has seen the review from servicem8addons_save_draft and said to deploy, and set confirm to "deploy". Waits for ServiceM8 (usually 20-40 seconds) and returns the result or the errors. Also turns a disabled add-on back on.

  • disable_addon write

    Switch an add-on off

    Switch a FloConnector add-on off: redeploy it with no buttons, menu entries or webhooks and a handler that does nothing. Takes effect in about 30 seconds. The draft is kept, so servicem8addons_deploy_addon turns it back on. Use this straight away if an add-on misbehaves. Removing an add-on from the account entirely is done by the user in ServiceM8.

  • preview_addon read

    Preview an add-on

    Show the user a clickable preview of a FloConnector add-on's saved draft: its job-card popup or Addons-menu page, run on sample data in their browser. Nothing reaches ServiceM8. Call it after servicem8addons_save_draft so the user can try the add-on before deciding to deploy. Only clients that show MCP Apps display it.

  • request_secret write

    Ask the user for an add-on secret

    Get a one-time FloConnector link where the USER pastes a secret value (an API key or token for another service) for a deployed FloConnector add-on. The value goes straight to ServiceM8 and never passes through this chat. Never ask the user to paste a secret into the chat, and never put one in the add-on's code. The link works once and expires in 30 minutes.

  • delete_secret write

    Delete an add-on secret

    Delete one secret from a FloConnector add-on. The add-on stops seeing it on its next run, so anything that uses it will fail until it is set again.

Questions, answered

How do I connect ServiceM8 Add-on Builder to Claude?

Create a FloConnector workspace, connect ServiceM8 Add-on Builder once from the dashboard, then add your FloConnector endpoint to Claude as a custom connector. Claude handles the sign-in prompt; there is no server to host and no code to write. The same endpoint works in ChatGPT.

What can Claude do in ServiceM8 Add-on Builder?

11 tools: 6 that read ServiceM8 Add-on Builder data and 5 that change it. Every tool is listed on this page, so you can see exactly what the model is able to call before you connect.

Does the AI see my ServiceM8 Add-on Builder password or API keys?

No. FloConnector holds the ServiceM8 Add-on Builder credential, encrypted before it is stored, and calls ServiceM8 Add-on Builder on the model's behalf. The AI client only ever sees the tools and their results, never the credential behind them.

Can I stop it changing anything in ServiceM8 Add-on Builder?

Yes. Each connection is exposed through a profile, and a profile carries only the tools you switch on. Leave the 5 write tools off and the connection is read-only.

Does this work with ChatGPT as well as Claude?

Yes. The endpoint is a standard hosted MCP server, so it works in any client that supports MCP connectors, including Claude and ChatGPT.