Arcade

Arcade is an authenticated tool-calling runtime for AI agents, built around a hard problem: letting an agent act on behalf of a specific user without ever exposing credentials to the model. It ships 7,500+ pre-built, agent-optimized tools across roughly 80 MCP servers (Gmail, Slack, GitHub, Salesforce, HubSpot, Stripe, and more), plus an open-source SDK for building custom tools. When an agent calls a tool, the Arcade Engine checks whether the calling user has authorized the required OAuth scopes, triggers the consent flow if not, and injects the token at runtime. The LLM and the client never see the secret.

It differentiates through identity-aware execution: per-user permissions enforced at the moment of execution, a permission-intersection model (agents act only within the overlap of agent and user scopes), integrations with Okta and Entra ID, and full audit trails of who called what, when, and whether it was authorized. LangChain partnered with Arcade in April 2026 to bring its tool library into LangSmith Fleet. The company raised a $12M seed led by Laude Ventures.

Arcade fits regulated teams and multi-user agents where every tool call must be authorized and auditable. Honest caveats: the catalog is smaller than Composio's or Paragon's; usage-based pricing ($0.01-$0.50 per execution) is hard to forecast at scale; it covers tool calls only, with no data syncs, webhooks, or triggers; and several governance features are enterprise-only.

Sign up at arcade.dev, authorize your tools per user via OAuth, and call them from any agent framework over MCP.