Run

MCP for AI agents

Floom exposes a Model Context Protocol server at https://floom.dev/mcp. Add it to Claude Desktop, Cursor, or any MCP-compatible client.

Add to Claude Desktop#

Claude Desktop config

// Claude Desktop config
// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "floom": {
      "url": "https://floom.dev/mcp",
      "headers": { "Authorization": "Bearer YOUR_AGENT_TOKEN" }
    }
  }
}

Agent tokens with run scope can run any owned private app. Tokens with publish scope can deploy via publish_app.

All 15 available tools#

Auth

auth_status	Check whether the current token is valid and what scopes it has.
start_device_flow	Begin CLI device-flow auth; returns a verification URL for the user.
poll_device_flow	Poll the device-flow until approved; returns an agent token on success.

Discovery

get_app_contract	Self-contained walkthrough for AI agents new to Floom.
list_app_templates	Browse scaffolding templates (e.g. multi_file_python, csv_stats).
get_app_template	Fetch a template by key and return its file map.
find_candidate_apps	Search for apps that match a natural-language task description.

Apps

list_apps	List all apps owned by the authenticated user.
get_app	Fetch metadata for a specific app by slug.
validate_manifest	Validate a floom.yaml before deploying (returns errors/warnings).
publish_app	Deploy an app from a file map (no local filesystem required).

Execution

run_app	Run any public or owned app by slug; returns output synchronously.
get_execution	Fetch the status and output of an async execution by ID.

Connections

list_my_connections	List the caller's active Composio OAuth connections.
set_secret	Set a secret value for an app (requires publish scope).

The get_app_contract tool returns a self-contained walkthrough designed for AI agents that need to understand how Floom works before building or running apps.

Example MCP call#

Example: run_app

# Claude calls run_app internally when you say:
# "Run my floom app csv-stats with this CSV..."
POST https://floom.dev/mcp
tool: run_app
arguments: { "slug": "csv-stats", "inputs": { "csv": "name,score\nAlice,90" } }