Automations, integrations & plugins

FlutterGo can work while you are away, and it can reach beyond your project. Automations are scheduled agent runs, "Scheduled automations that run unattended," as the page puts it. Integrations connect external systems: MCP servers add third-party tools to your agent, and connectors link accounts like Supabase, GitHub, Notion, or Linear.

Both live in the left sidebar: Automations and Integrations. Here we cover each, plus how connected resources show up in the chat composer.

The Integrations page describes its goal in one line: "Connect external systems, bring MCP tools into your agent loop, and use FlutterGo AI from other IDEs, scripts, and automations."

What you need#

• You are signed in, and the sidebar is visible (click Expand sidebar if it is collapsed).
• For automations that reuse a project: that project already exists.
• For connectors on a self-hosted instance: your administrator has enabled connectors (they need an instance Composio API key).

How to do it#

Part A, Automations#

1. Open Automations#

Click Automations in the sidebar. The page is titled Automations, "Scheduled automations that run unattended." If it is empty you'll see "No automations yet." and the hint "Click New automation to schedule an unattended agent run."

2. Click New automation and fill in the basics#

Click New automation. Fill in:

• Name, e.g. the placeholder suggestion, "Morning briefing".
• Prompt, what the agent should do each run. Example placeholder: "Pull yesterday's GitHub + Linear activity and summarize what changed."

3. Pick a schedule#

Under Schedule, choose one of: Hourly, Daily, Weekdays, Weekly. Depending on your choice you set Minute of every hour, or a Time (AM/PM) and Timezone, and for weekly runs a day. The form describes the result in plain words, e.g. "Runs daily at {time} {tz}" or "Runs Mon–Fri at {time} {tz}".

4. Choose where each run works#

Under Project, pick one of two modes:

• Create a new project each run, "A fresh, isolated workspace per fire."
• Reuse an existing project, "Each run lives as a new conversation inside the project." Pick the project from the list ("— Pick a project —").

Click Create to save. (Editing later uses Save.)

Important: If you choose reuse but skip the picker, you'll see "Pick a project to reuse, or switch to 'Create new each run'".

5. Run, watch, and manage#

Each automation card offers:

• Run now, fire it immediately (runs are marked manual vs scheduled).
• History / Hide history, past runs with statuses: queued, running, succeeded, failed, canceled. Use Open project to jump to "the project this run wrote to".
• Pause / Resume, paused automations show a paused tag; the card also shows "next: {when}" and the last run.
• Edit and Delete, deleting asks: "Delete this automation? Past runs and their projects are kept."

Success: A morning automation with "Create a new project each run" gives you a fresh, ready-made project in your Projects list every day.

Part B, Integrations#

6. Open Integrations and pick a tab#

Click Integrations in the sidebar. The view has tabs: MCP ("External tools"), Connectors ("Accounts and APIs"), and Skills.

7. Add an MCP server#

The MCP tab is titled External MCP servers, "Third-party tools for your coding agent." Click Add server. As the empty state says: "pick a template (Higgsfield OpenClaw, Pollinations, Allyson, Imagician, EdgeOne Pages, GitHub, Filesystem…) or set up a custom stdio / HTTP server." Click Save changes when done; the config location is shown under Stored at.

Note the forwarding rules shown on the page: some agents receive these servers automatically (Forwarded to:), others don't (Not forwarded to:), for those, configure MCP in the agent's own config file (e.g. ~/.codex/config.toml).

8. Connect a connector#

The Connectors tab lists integrations by Category (CRM, Database, Communication, and many more), with a search box ("Search connectors…"). Each card shows a status: Available, Connected, Error, or Disabled, plus a tools badge ("{n} tools").

1. Open a connector's details and click Connect.
2. An authorization window opens, the panel shows "Waiting for authorization..." ("Complete authorization in the opened window."). Click Cancel to abort.
3. Once Connected, the Tools section lists what the agent can now do. Use Disconnect to remove access.

9. The Skills tab#

The Skills tab (Skills integrations) is currently a reserved surface: "Skill-level integration management is being carried over from another branch. This tab is reserved so MCP, Connectors, and future Skills setup live in the same Integration route." Skills themselves already run automatically in every project, see Prompt, design & code generation.

10. Use everything from the chat composer#

Connected resources are available where you prompt. In the composer's add-context menu you'll find Add plugin, Add connectors, and Add MCP server (with empty states like "No MCP servers" or "No connected connectors" until you set them up). You can also type @ to mention a specific skill, plugin, MCP server, or connector in a prompt.

Tip: Once a connector is connected, just ask in plain language, e.g. with Supabase connected: "add sign-in with Supabase". The agent uses the connector's tools itself.

Common mistake: Adding an MCP server and expecting it in every agent. Check the Forwarded to / Not forwarded to notes on the MCP tab, some CLIs need MCP configured in their own config file.

Troubleshooting#

Problem: The MCP tab can't save or load. Possible causes: The local daemon is not running. Solution:

1. The page will say "Could not reach the local daemon. Make sure FlutterGo AI is running, then reopen this panel."
2. Start the FlutterGo AI app, then reopen Integrations.
3. If saving fails: "Save failed. Check that the daemon is running and try again."

Problem: Connectors ask for an API key. Possible causes: On self-hosted instances, connectors need a Composio key ("Add your Composio API key to continue"). Solution:

1. If you administer the instance: paste the key and click Save key ("Paste your key above and click Save key to load available integrations.").
2. Otherwise, ask your administrator to enable connectors.

Problem: Authorization never completes. Possible causes: The OAuth pop-up was blocked or closed. Solution:

1. Click Cancel, allow pop-ups for the studio, and click Connect again.
2. Complete the sign-in in the opened window before returning.

Problem: An automation run "succeeded" but produced nothing. Possible causes: The model returned an empty response ("Agent completed without producing any output…"). Solution:

1. Open the run's project and check the conversation.
2. Make the prompt more concrete, then Run now to test.

FAQ#

Do automations run when my computer is off? Automations are scheduled server-side runs designed to fire unattended at their scheduled time; check each automation's "next:" line and run History to confirm firings.

What is the difference between a connector and an MCP server? A connector links an account (Supabase, Notion, Linear…) through managed OAuth and exposes its tools. An MCP server is a tool server you add yourself, a template or a custom stdio/HTTP server.

Can I test an automation without waiting for the schedule? Yes, click Run now on its card. The run is tagged manual in history.

Are old runs deleted with the automation? No. The delete confirmation says it: "Past runs and their projects are kept."

Where do I manage skills? Skills run automatically inside projects. The Integrations Skills tab is reserved for future skill-level management; instance administrators manage the skill catalogue in the admin console.

Can the agent in a normal chat use my connectors? Yes. Once connected, connector tools are part of the agent's toolbox, and you can attach one explicitly via Add connectors or an @-mention.

Where to go next#

• Working with the AI agent, mentioning skills, plugins, and connectors in prompts.
• Deploying your app, store deployment uses connectors too (Play service account, App Store key).
• Save your code to GitHub, GitHub is its own first-class integration.