Local MCP server

Run Fload as a local MCP server with an API key. It's for AI clients that can't connect to a remote server, and for scripts, CI and other places where nobody is around to sign in. The package passes every request on to Fload's remote server, so you get the same tools as an AI app that signs in, within your key's permissions.

Most people should use the remote server instead. You sign in once, with no key to manage, and get every tool. It also takes an API key if you'd rather not run anything locally. See Connect via MCP.


Set it up in three steps

  1. 1

    Create a key

    In Fload, Settings › Connect (MCP). See API keys for what each permission allows.

  2. 2

    Add the config

    Paste the block below into your client's MCP config, with your key.

  3. 3

    Restart your client

    It starts the server on your machine, and the server talks to Fload with your key.

You need Node.js, so your client can run npx.


The config

Add this to your client's MCP config (for Claude Desktop that's claude_desktop_config.json; Cursor and other clients use the same block):

json
{
  "mcpServers": {
    "fload": {
      "command": "npx",
      "args": ["-y", "@fload-ai/mcp"],
      "env": {
        "FLOAD_API_KEY": "fload_sk_your_key_here"
      }
    }
  }
}

Restart your client. It starts the package on your machine, and the package talks to Fload's remote server with your key.

In Claude Code, one command does it:

bash
claude mcp add --env FLOAD_API_KEY=fload_sk_your_key_here --transport stdio \
  --scope user fload -- npx -y @fload-ai/mcp

Prefer not to put the key in your client config? Save it in ~/.fload/config.json instead:

json
{
  "apiKey": "fload_sk_your_key_here"
}

The package is @fload-ai/mcp on npm. This page describes version 0.13.0 and later.


Tools

The same as the remote server: see every tool under Connect via MCP. A tool your key has no permission for answers with the permission it needs.

Moving from 0.12

Version 0.13.0 replaced the package's own tools with the server's. Your agent keeps working:

  • The old tool names still work for two more versions, each passing to the tool that now does the work. Switch to the new names when you can:
  • list_appslistAssets
  • get_app_detailsgetAppDetails
  • get_reviewsgetReviews
  • generate_review_replygenerateReviewReply
  • send_review_replyreview_stage_reply
  • translate_reviewtranslateReview
  • discover_metricsdiscoverMetrics
  • get_metricsgetMetrics
  • discover_dimensionsmetric_dimensions
  • get_download_impactgetDownloadImpact
  • list_agentsgetAgentStatus
  • get_agent_detailsgetAgentStatus
  • get_agent_run_historyagent_run_history
  • trigger_agent_runtriggerAgentRun
  • pause_agentpauseAgent
  • resume_agentresumeAgent
  • get_agent_activityagent_activity
  • get_ads_performanceads_campaigns
  • get_dashboard_overviewgetDashboardOverview
  • list_pending_actionsgetTaskLedger
  • approve_actionapprovePendingAction
  • reject_actionrejectPendingAction
  • check_action_nowcommandActionItems
  • rename_custom_page_actioncommandActionItems
  • get_aso_summarygetAsoSummary
  • get_aso_recommendationsaso_recommendations
  • get_aso_evidenceaso_get_evidence
  • get_aso_keywordsaso_tracked_keywords
  • get_aso_experimentsaso_experiments
  • get_aso_locale_snapshotsgetAsoListingSnapshots
  • trigger_aso_analysisaso_refetch_listing
  • get_conversation_messagesget_conversation
  • Review replies wait for approval. send_review_reply now puts the reply in your inbox instead of sending it. Nothing reaches the store until the reply is approved.
  • No more chat messages to Fio. send_chat_message is gone, since your AI app already has Fload's tools. list_conversations and get_conversation still read your chats with Fio.
  • Three tools need the app's id. list_agents, list_pending_actions and get_ads_performance now take assetId, from listAssets.

Good to know

  • One organization per key. A key reaches every app in the organization it was made in, and acts as the person who made it.
  • A key never does more than your role allows.
  • Changes still go through your inbox. A change requested with an API key always waits there first. It's never applied straight away.
  • Some things need a person. An API key can't turn an agent on, let an agent act on its own (such as ASO auto mode), or approve publishing screenshots. See API keys.
  • Credits: reading data is free. AI work, like drafting a reply, uses the same credits as in the app.
  • Different API address? Set FLOAD_API_URL. The default is https://api.fload.com.