Arcade
Also known as: Arcade AI
Authenticated tool calling platform and MCP runtime that handles OAuth and token management so AI agents can securely act across 7,000+ integrations.
Arcade is an authenticated tool calling platform and Model Context Protocol runtime for production AI agents. It was founded in 2024 by a team of former Okta identity engineers, including Alex Salazar, Sam Partee, and Atul Tulshibagwale, and its core idea is often described as single sign on for AI. The platform lets an agent take real actions in outside services, such as sending an email, creating a GitHub issue, or updating a CRM record, while handling the OAuth flows, token storage, and scope management that those actions require.
The defining design choice is that the language model never touches a credential. When an agent calls a tool, Arcade authorizes the user just in time, keeps the token in its runtime, and returns only the result, so secrets never enter the reasoning loop. Users are challenged for new permissions by scope, and enterprises can layer their own policy on top, for example enforcing a Cisco Duo push the moment an agent reaches for sensitive data. This auth first posture is what Arcade sells to security reviewers who would otherwise block agents from production.
For developers, Arcade ships more than 7,000 prebuilt integrations for services like Gmail, Slack, GitHub, and Salesforce, plus a Python SDK for building and testing custom tools and a registry for publishing them. It exposes an OpenAI compatible endpoint, so teams already using function calling can point requests at Arcade and gain authenticated tools without rewriting code, and it integrates with LangChain, CrewAI, OpenAI Agents, and Vercel AI. Arcade positions itself as an MCP runtime, builds MCP servers for clients including Cursor, VS Code, and Claude Desktop, and is a Gold Member of the Linux Foundation Agentic AI Foundation.
Arcade deploys across cloud, private VPC, and on premises, and the engine can also run locally, which suits regulated buyers in financial services and healthcare. Pricing is usage based: a permanent free tier covers light experimentation, a Growth plan adds a $25 monthly platform fee with included allocations, and usage beyond that is billed per user challenge and per tool execution. The company raised a seed round of about $12M led by Laude Ventures, with a larger Series A reported since, and it remains independent.
Vendor details
Canonical URL
https://arcade.dev
Category
Agent infrastructure
Subcategory
Agent auth and tool calling
Funding status
Founded 2024 by former Okta identity engineers (Alex Salazar, Sam Partee, Atul Tulshibagwale). Seed of about $12M led by Laude Ventures, with a larger Series A reported since.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
More than 7,000 prebuilt tool integrations (Gmail, Slack, GitHub, Salesforce, and more), a Python SDK for custom tools, an OpenAI compatible endpoint, MCP servers, and connectors for LangChain, CrewAI, OpenAI Agents, and Vercel AI.
In practice
You want your agent to act in Gmail and Salesforce for a user without the model touching OAuth tokens. Arcade authorizes the user just in time and runs the tool, keeping credentials out of the prompt.
Your security team blocks the agent because no one can tell who authorized which action. You route tool calls through Arcade, which enforces per user scopes and your existing policy and gives reviewers an authorization trail.
You keep rebuilding OAuth and token handling for every new integration. You point requests at Arcade's OpenAI compatible endpoint, pull from 7,000+ prebuilt tools, and add custom ones with the SDK instead of building auth yourself.
Sources & related URLs
Related / legacy domains
Capability coverage
8.0 / 14 capabilities · 57%
| Integrations & Tool CallingCore product. More than 7,000 prebuilt tool integrations across Gmail, Slack, GitHub, Salesforce, and others, with authenticated tool calling as the central function. | Full |
|---|---|
| Workflow OrchestrationHandles tool execution, parallel calls, and error handling at the tool layer, but defers workflow sequencing and branching to agent frameworks like LangChain and CrewAI. | Partial |
| Knowledge Grounding & RAGNo retrieval or RAG grounding layer. Arcade provides authenticated tools that can fetch data, but knowledge grounding is not part of the platform. | Unable to verify |
| Human Oversight & GuardrailsStrong authorization guardrails: scope based user challenges, just in time consent, URL elicitation, and enterprise policy enforcement such as a Cisco Duo push on sensitive actions. Scoped to permissions rather than a general approval workflow engine. | Partial |
| Security, Identity & GovernanceDefining strength. Built by former Okta engineers; OAuth and token management with credentials kept out of the model, scope based authorization, and enterprise identity and policy controls. | Full |
| Observability & AuditabilityObservability is a stated core pillar with a usage dashboard and tool call monitoring, but published detail on full audit log export and tracing depth is limited. | Partial |
| Memory & State PersistencePersists user authorization state and tokens across sessions but provides no agent memory or conversation state layer. | Unable to verify |
| Deployment & Data ResidencyDeploys across cloud, private VPC, and on premises, and the engine can run locally, supporting data residency for regulated buyers. | Full |
| Prebuilt Agents, Templates & PacksCatalog of more than 7,000 prebuilt integrations plus a registry for publishing tools and sample agents, which sharply reduces time to value. | Full |
| Triggers & Channel CoverageSurfaces through MCP clients such as Cursor, VS Code, and Claude Desktop and through agent frameworks, but provides no native scheduling, webhooks, or conversational channels of its own. | Partial |
| Model Flexibility & RoutingModel agnostic through an OpenAI compatible endpoint that works with any provider, but Arcade does not itself route or select between models. | Partial |
| APIs, SDKs & MCP ExtensibilityOpenAI compatible API, Python SDK for custom tools, and a first class MCP runtime that builds MCP servers for any client. | Full |
| Testing, Debugging & OptimizationProvides tooling to benchmark and test LLM tool interactions before production, scoped to tool calls rather than full agent output quality scoring. | Partial |
| Browser & Computer UseAPI and tool based. No browser automation or computer use; actions are taken through authenticated APIs rather than GUI control. | Unable to verify |
Pricing
From $25/mo · free tier
Platform fee plus usage (user challenges and tool executions)
Included quota
Free tier includes 100 user challenges, 1,000 standard tool executions, 50 pro tool executions, and unlimited tools with prebuilt auth. Growth ($25/mo) includes 600 user challenges, 2,000 standard tool executions, and 100 pro tool executions. Rate limit is 1,000 API calls per minute on all tiers.
What is public
Arcade publishes usage based pricing: a permanent free tier, a Growth plan at $25 per month with included allocations, and per-resource overage rates. Enterprise limits are by contact.
Billing mechanics
Plans bundle monthly allocations of user challenges and tool executions. The $25 Growth platform fee is billed monthly; usage beyond the included allocations is metered and billed at the end of the month at $0.05 per user challenge, $0.01 per standard tool execution, and $0.50 per pro tool execution. A user challenge is a prompt for new permissions; a tool execution is one agent tool call.
Cost watchouts
Pro tool executions at $0.50 each and high volume agents drive overage. Allocations do not appear to roll over month to month.
Variable cost rationale
Included allocations are generous and standard tool executions are cheap at $0.01 each, so light and mid usage is predictable. Cost grows with high volume production agents, and pro tool executions at $0.50 each can add up.
Additional watchouts
Cost scales with tool execution volume, which is hard to predict for high traffic production agents. Pro tool executions are priced fifty times higher than standard ones.
Overage / add-ons
Beyond included allocations, Growth bills $0.05 per user challenge, $0.01 per standard tool execution, and $0.50 per pro tool execution, charged monthly at the end of the cycle.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free tier: 100 user challenges, 1,000 standard tool executions, 50 pro tool executions per month, unlimited tools
Lowest paid plan
Growth: $25/mo platform fee plus usage
Commercial notes
Arcade is sold to engineering teams building production agents and to enterprises with strict security review. Its differentiator is auth: built by former Okta engineers, credentials never reach the model. Deploys to cloud, VPC, on premises, and locally.
Key ambiguities
Total cost depends on production tool call volume and the mix of standard versus pro executions, which is hard to forecast before deployment.
Cancellation / refund
Self serve plans are month to month and usage is billed in arrears. Cancellation details for paid plans are not publicly detailed.
Support SLA / resale
Self serve support for free and Growth tiers; enterprise terms and higher rate limits are by contact at contact@arcade.dev.
Missing data
Enterprise pricing is not public. The most recent public pricing iteration dates to 2025, so current rates should be confirmed on arcade.dev.
Related vendors
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- Apify — Cloud platform for web scraping and automation with 45,000+ prebuilt…
- Arize AI — AI observability and evaluation platform that traces, evaluates, and…
- Braintrust — AI evaluation and observability platform with self-serve pricing,…
- CalypsoAI — Enterprise AI security platform that red teams, defends, and…