Arcade

What it does

Arcade positions itself as the enterprise-ready actions runtime for AI agents. Its docs break the problem into authorization, execution, and lifecycle governance. Arcade handles OAuth and manages user tokens and secrets, enforces identity provider, DLP, SIEM, and compliance policies plus per-action authorization at runtime, and provides shared registry, version control, visibility filtering, and OpenTelemetry audit logs for governing agents across an organization.

The execution catalog is MCP native. The docs describe 7,500+ agent-optimized tools across 81 servers, and the arcade-mcp README gives the same catalog figures for prebuilt tools and servers. If a needed tool is not in the catalog, teams can register a remote MCP server they or a vendor already runs, or build custom tools with the open-source arcade-mcp Python framework. That framework provides a decorator API for MCP tools, resources, prompts, sampling, elicitation, progress, and logging, plus authorized tool calling where a tool declares required OAuth scopes and Arcade handles prompting, token storage and refresh, and per-call scoping so the client and LLM never see the token.

Arcade can run locally for single-user agents and on the hosted platform for multi-user production agents. Custom local tools and custom hosted tools share tool code, the platform zone provides project and user management, auditing, tool registry, authentication and secrets, and a distributed runtime, and external MCP servers connect through the same engine. Pricing has a free tier for prototyping, a Team plan with a platform fee plus usage for auth events and tool calls, and Enterprise terms with VPC or air-gapped deployment, SSO, RBAC, audit logs, and private registry access.

Verified capabilities

Quick start

Scaffold a custom Arcade MCP server

Install the arcade-mcp CLI, scaffold a server, and define a minimal tool. This follows the arcade-mcp README quickstart for builders who need a tool that is not already in the catalog.

uv tool install arcade-mcp
arcade new my_server

from typing import Annotated
from arcade_mcp_server import MCPApp

app = MCPApp(name="my_server", version="1.0.0")

@app.tool
def greet(name: Annotated[str, "Name to greet"]) -> str:
    """Greet a person by name."""
    return f"Hello, {name}!"

if __name__ == "__main__":
    app.run(transport="stdio")

Source: https://github.com/ArcadeAI/arcade-mcp. Examples use placeholders only. Never paste a real key into a profile, config file you share, or a ticket.

MCP support: Current MCP support details were not cleanly verified for Toolhouse in this research pass. Check the official Toolhouse documentation for the supported setup before wiring an MCP client to it.

Works with

Only sourced support is listed. A missing framework means AgentsUse has not verified it yet, not that it cannot work.

Health and maintenance

Pricing and license

Free tier to prototype on Arcade Cloud, Team plan with a monthly platform fee plus usage charges for auth events and tool calls, and custom Enterprise pricing with annual bundles and deployment options. See the pricing page for current rates.

Official pricing or docs →

Limitations and safety

• The free tier is small for production use, with monthly auth event and tool call allowances shown on pricing, so chatty agents can move to paid usage quickly.
• Catalog coverage matters. Docs encourage checking prebuilt tools first and building or registering custom MCP servers for internal APIs, custom OAuth providers, or missing integrations.
• The open-source MIT piece verified here is the arcade-mcp framework. The hosted Arcade Engine, registry, and governance plane are commercial platform services with Enterprise controls on paid terms.

Alternatives to Arcade

Related tools

Common questions