Arcade.dev: The Actions Runtime for AI Agents, Explained
Arcade.dev is infrastructure that sits between AI agents and the business apps they act in. When an agent needs to send an email, update a CRM record or file a ticket for a specific person, Arcade checks that this agent, acting for this user, is allowed to take that action, injects the user’s credentials without ever showing them to the AI model, runs the action, and logs who did what. It calls itself “the actions runtime between your agents and every system they reach,” and offers more than 8,000 pre-built tools plus an enterprise MCP gateway. Arcade AI Inc. was founded in 2024 by Alex Salazar and Sam Partee and raised a $60 million Series A in June 2026.
The quick answer: Arcade is built for companies getting AI agents past their security review. Its specialty is delegated, per-action authorization: the agent acts as its user, never beyond that user’s permissions or its own scope, with the rules enforced outside the model where a prompt can’t undo them. It’s free to start and priced by tool calls rather than seats. A disclosure: we make Carly, an AI assistant that connects to thousands of apps for its users, so we work on a related problem from the end-user side. Everything below comes from Arcade’s site, docs and funding coverage.
Arcade at a glance
Checked against Arcade’s site, documentation, pricing and funding coverage on November 12, 2026.
| Arcade.dev | |
|---|---|
| What it is | An authorization and execution layer for AI agent actions, plus an MCP gateway |
| Tools | 8,000+ permission-aware MCP tools, plus custom MCP servers you build or bring |
| Works with | Claude Code, Cursor, Codex, GitHub Copilot, Claude, and frameworks like LangChain and CrewAI |
| Deployment | Arcade Cloud, your own cloud, on-premises or air-gapped |
| Pricing | Free with 2,000 tool calls a month; Team at $25 a month plus $0.01 per call; Enterprise custom |
| Compliance | SOC 2, SSO, role-based access and audit logs |
| Company | Arcade AI Inc.; founded 2024; $72M raised |
What Arcade does
Arcade’s homepage frames it as “one control point to enforce, execute, and govern every action, every time.” Its docs say it’s needed when an agent has to act in a system that belongs to a real user or company, such as sending an email, updating a CRM record, filing a ticket or moving money, “and someone will be asked who authorized it.”
Arcade organizes the product around three jobs:
- Enforce. Authorization for every action, evaluated at runtime against your identity provider, “at the intersection of what the agent AND the user are allowed to do,” plus the security policies a company already runs, such as data loss prevention, applied through hooks.
- Execute. A large catalog of tools built for agents rather than thin wrappers around APIs, so a request like “make the intro paragraph friendlier” maps to the right document segment, position and text.
- Govern. One control plane for every agent and tool in a company: a shared registry, version control for tools, filtering so agents only see tools their user may use, and audit logs exported to security tools over OpenTelemetry.
How authorization works
Arcade’s core idea is delegated authorization. Instead of giving an agent a broad service account or a hard-coded token, the agent acts on behalf of an authenticated user. “Every action runs at the intersection of what the user can do and what the agent is scoped to do. Cross either line and the action stops.”
The flow, according to Arcade’s developer guide, works like this:
- Discover tools. The agent lists the tools available to it.
- Check permissions. Arcade reports which authorization requirements are missing, down to the exact OAuth scopes a provider still needs.
- Get consent. Arcade returns a link to send the user to, and waits until they grant access.
- Act. The agent executes the tool, and Arcade injects the provider’s token at that moment. The token “is never returned to the caller or the model.”
That last point is central to its security pitch: credentials never sit in the model’s context, where a malicious prompt could try to extract them. SiliconANGLE reported that Arcade encrypts tokens before storing them and adds salting as an extra safeguard.
Tools and the MCP gateway
Arcade offers more than 8,000 “permission-aware” tools for the apps companies run on, such as Google, Slack and Salesforce. Two features aim at agents with large toolsets:
- Tool recommendation gives an agent the whole catalog while cutting token costs by up to 37%, according to Arcade.
- Built-in evals catch hallucinated and destructive tool calls before they run.
Arcade also works as an MCP gateway, federating many MCP servers behind one authenticated URL. Developers can build their own MCP servers with its open-source arcade-mcp framework, which has about 1,100 stars on GitHub, and deploy them to Arcade with managed authorization.
Getting started is quick. Arcade publishes plugins and install steps for Claude Code, Cursor, Codex, GitHub Copilot and Claude, and its framework guides cover LangChain, LlamaIndex, CrewAI, Mastra, Pydantic AI, Google ADK, OpenAI Agents and Vercel AI.
When Arcade isn’t the right tool
Arcade’s own docs are clear about what it isn’t for:
- You need a model, an inference endpoint or a vector database.
- Your agent only reads public data and never touches a user’s account.
- You want an agent-building framework. Arcade is framework agnostic.
- You want “a hosted chat product for end users rather than infrastructure for the agents you build.”
Pricing
Arcade’s pricing charges for what agents do, not per seat:
| Plan | Price | What it includes |
|---|---|---|
| Free | $0 | 2,000 tool calls a month on Arcade Cloud, community support, no credit card |
| Team | $25 a month plus usage | Tool calls at $0.01 each, next-business-day email support |
| Enterprise | Custom | Annual call bundles, VPC or air-gapped deployment, 24/7 support, SSO, RBAC and audit logs |
The company
Arcade AI Inc. was founded in 2024 by Alex Salazar, its CEO, a former Okta vice president who built products there, and Sam Partee, its CTO, previously at Redis. TechCrunch reported that the pair started out building a site reliability agent, but kept “beating our heads against the wall” getting it to connect to other services, which led to Arcade. The company raised a $12 million seed round from Laude Ventures in March 2025.
In June 2026, SiliconANGLE reported a $60 million Series A led by SYN Ventures, with Morgan Stanley and Wipro joining. “Agents don’t fail in production because the model is wrong,” Salazar said. “They fail because nobody can prove that for any given action by an agent, whether that agent on behalf of that user can perform that action on that resource.”
Arcade says its team comes from Okta, Snowflake, Redis, Airbyte and MongoDB, and that it authored the MCP tool authorization specification and sits on steering committees for MCP security and governance.
How Arcade compares
Arcade competes with other tool and integration platforms for AI agents, including Composio, Nango, Paragon and Pipedream Connect. Our Composio alternatives guide compares them. Arcade’s distinguishing focus is enterprise governance: per-action authorization tied to a company’s identity provider, audit logs and air-gapped deployment, where Nango leans toward data syncs and integration code you own.
Arcade is infrastructure for teams building agents. If you want an assistant that’s already connected to the apps you use, Carly connects to thousands of apps, labels and sorts email across all your Gmail and Outlook accounts by your rules, handles advanced scheduling, and makes and answers calls and texts, from inside ChatGPT or Claude or by text.
Who Arcade suits
Arcade is a good fit if:
- You’re building agents that act in your users’ or employees’ accounts, and someone will ask who authorized each action.
- Security or compliance review is what’s blocking your agent from production.
- You need audit logs in your SIEM, or to deploy in your own cloud or air-gapped.
- You want one gateway in front of many MCP servers.
Look elsewhere if:
- Your agent only reads public data.
- You need data syncs and webhooks more than agent actions.
- You want a finished assistant rather than infrastructure to build one.
FAQ
What is Arcade.dev?
Arcade.dev is an actions runtime for AI agents. It sits between agents and the apps they act in, handling per-user authorization, OAuth tokens, a catalog of 8,000+ tools, an MCP gateway and audit logs.
How much does Arcade cost?
Arcade is free for 2,000 tool calls a month. The Team plan is $25 a month plus $0.01 per tool call, and Enterprise is custom.
Does Arcade show credentials to the AI model?
No. Arcade injects the provider’s token when a tool runs, and says the token is never returned to the caller or the model.
Does Arcade work with Claude Code and Cursor?
Yes. Arcade publishes plugins and setup steps for Claude Code, Cursor, Codex, GitHub Copilot and Claude, plus a remote MCP server URL for other clients.
Who founded Arcade?
Alex Salazar, formerly at Okta, and Sam Partee, formerly at Redis, founded Arcade in 2024. It has raised $72 million, including a $60 million Series A in June 2026.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
