What is
Arcade
?
Most platform initiatives fail the same way: they start too big, get funded like projects instead of products, build for too long without testing with users, and confuse renaming an infrastructure team with building a platform team. Instead, start small, learn fast and scale deliberately.
At a glance
Fact | Value |
|---|---|
Primary use case | Secure execution and centralized governance of AI agent actions across business systems |
Deployment type | Cloud, on-premises, air-gapped, or hybrid; Helm chart available for Kubernetes deployment |
Pricing model | Free tier and paid offerings; complete paid pricing not specified |
API compatibility | Model Context Protocol (MCP), OAuth |
Supported integrations | Google, Slack, Salesforce, Twitter, LinkedIn, GitHub, identity providers, policy engines, SIEM, observability and evaluation tools, custom MCP tools |
Source type | Dual |
Cloud provider | Cloud Agnostic |
Ideal team size | Enterprise teams; no minimum size specified |
Required skills | AI agent integration, MCP, authentication and delegated authorization, security policy configuration; Kubernetes and Helm for self-hosted deployment |
AEP branch | Agent Infrastructure |
AEP category | Harness |
AEP subcategory | Capability |
AEP detail | Tool Gateways |
Profile
Tool Gateways provide a controlled point for connecting AI agents to tools and business systems. Arcade is an actions runtime that authenticates users through their identity provider, delegates authorization so agents act within the user's scope, and applies policies to each action outside the model.
The platform provides more than 8,000 permission-aware tools and supports custom MCP tools. Its control plane manages tool access and versions, and action records identify the agent, user, and system; records can be streamed to a SIEM. Arcade also offers deployment in cloud, on-premises, air-gapped, or hybrid environments.
Focus
Arcade is intended for enterprise teams that need to put AI agents to work across business systems while retaining centralized security and governance controls. It addresses risks associated with agents using shared service accounts, inheriting broad user permissions, or relying on separately built security infrastructure by enforcing delegated authorization and policy in the action path.
Background
Arcade says its team comes from Okta, Snowflake, Redis, Airbyte, and MongoDB. The company says it authored the MCP tool authorization specification and participates in steering committees for MCP security and governance. The supplied research does not specify Arcade's founding date or funding history.
The company background research reports that most Arcade MCP servers were removed from the ArcadeAI/arcade-ai repository and transitioned to closed source. The research does not provide further details about the company's open-source contribution history.
Features
Authentication and delegated authorization
Agents can take actions on behalf of users without receiving permissions beyond the user's authorization scope. Arcade authenticates through the user's identity provider and applies existing policies in the action path, outside the model.
Permission-aware tool execution
Agents can use a catalog of more than 8,000 permission-aware tools, along with custom MCP tools. Built-in evaluations check for hallucinated or destructive calls before execution.
Centralized governance and audit records
Teams can manage tool access from one control plane, including granting, revoking, and versioning tools. Each action record names the agent, user, and system, and can be streamed to a SIEM.
Deployment and integration options
Arcade supports cloud, on-premises, air-gapped, and hybrid deployments, and works with different clients, models, and frameworks. It can run alongside existing identity providers, policy engines, observability tools, and evaluation tools.
Enterprise access controls
Arcade lists SOC 2 compliance, SSO, role-based access control, and full audit logs among its security features. These controls provide identity and activity-management capabilities for teams operating agents.
Pros and cons
Pros
Delegated authorization lets agents act as users without exceeding their permissions, and policies are enforced outside the model.
More than 8,000 permission-aware tools are available, with support for custom MCP tools.
Centralized tool management and records that identify the agent, user, and system support governance and auditability.
Cloud, on-premises, air-gapped, and hybrid deployment options are available.
Cons
The complete paid pricing is not specified in the supplied research, so buyers cannot determine full paid-plan costs from the available pricing details.
Most Arcade MCP servers were removed from the public ArcadeAI/arcade-ai repository and transitioned to closed source, limiting public access to those server implementations.
Self-hosted Kubernetes deployment requires Kubernetes and Helm skills, in addition to AI agent integration, MCP, authentication, delegated authorization, and security policy configuration skills.
FAQs
Arcade FAQ
How does Arcade govern AI agent actions?
Arcade authenticates users through their identity provider, delegates authorization so agents act within the user's scope, and enforces policies in the action path outside the model. Each action record identifies the agent, user, and system, and can be streamed to a SIEM.
Which integrations does Arcade support?
Arcade supports Google, Slack, Salesforce, Twitter, LinkedIn, GitHub, identity providers, policy engines, SIEM, observability and evaluation tools, and custom MCP tools.
How much does Arcade cost?
Arcade offers a free tier that includes 2,000 authentication events and 2,000 tool calls monthly; paid offerings are also available, but complete paid pricing is not specified in the supplied research.
Where can Arcade be deployed?
Arcade supports cloud, on-premises, air-gapped, or hybrid deployment, and a Helm chart is available for Kubernetes deployment.
What skills are required to deploy Arcade?
Arcade requires AI agent integration, MCP, authentication and delegated authorization, security policy configuration; Kubernetes and Helm are required for self-hosted deployment.
What are Arcade's limitations?
Arcade's complete paid pricing is not specified in the supplied research, and most Arcade MCP servers were removed from the public ArcadeAI/arcade-ai repository and transitioned to closed source.
