One (withone.ai) Review: The AI Agent Integration Runtime
7 min read
Updated
A detailed review of One (withone.ai), an integration runtime giving AI agents authenticated access to 250+ platforms through a single CLI and MCP server.
If you have built an AI agent that needs to interact with multiple applications, you are likely familiar with the integration challenges. Every API requires its own OAuth flow, token refresh logic, error handling, and rate limit management. Managing these across dozens of applications can divert significant time away from core agent development.
One (formerly Pica) aims to resolve this problem. It is an integration runtime designed to give agents authenticated access to over 250 platforms via a single CLI, dashboard, and universal Model Context Protocol (MCP) server. By connecting your applications once, your agent can perform actions across Gmail, Slack, Stripe, GitHub, HubSpot, and other services without needing custom OAuth implementations.
What One Does
One handles authentication, token refreshes, rate limits, and API schemas so developers can focus on their agent's core decision-making logic rather than endpoint integration details.
The primary components include:
- CLI (
one): A terminal interface to search for integrations, connect applications, and execute actions. - Universal MCP Server: A hosted MCP server offering over 47,500 vetted actions, compatible with Cursor, Claude Desktop, and other MCP-supporting clients.
- One Flow: A workflow runtime for multi-step agent tasks. It supports branching, loops, parallel execution, and pause/resume capabilities, allowing agents to generate and modify flows as JSON.
- AuthKit: An embeddable authentication interface designed to let end-users securely connect their own accounts, with silent token refreshes and isolated credentials.
- Bridge: A utility that generates a hosted MCP server for any niche or internal API when provided with its documentation URL.
The SDK, skill definitions, and MCP tools are open-source and publicly accessible.
Pricing Breakdown
One offers a tiered pricing structure starting with a free tier. Every plan includes 1 million API calls.
| Plan | Cost | API Calls | Rate Limit | Key Features |
|---|---|---|---|---|
| Free | $0 | 1M included | 100/min | Unlimited connections, basic relay (5), 14-day logs |
| Starter | $29/mo | 1M + $0.25/1K overage | 1K/min | 50-app AuthKit, 100 relay, email support |
| Pro | $199/mo | 1M + $0.15/1K overage | 10K/min | Unlimited AuthKit/relay, 90-day logs, Slack support |
| Enterprise | Custom | Custom | Custom | SSO, SCIM, dedicated SLA, granular access |
The free tier requires no credit card and provides full CLI and MCP access alongside unlimited connections. Key limitations of the free tier include a rate limit of 100 calls per minute, a maximum of 5 relay endpoints, and a 14-day log retention period.
For production systems, overage fees apply once the monthly 1 million call limit is exceeded, starting at $0.25 per 1,000 additional calls on the Starter plan.
The CLI Experience
The one CLI serves as the primary developer interface. Running one connect initiates application authentication via the browser. Once connected, users can search for specific actions (e.g., one search "send email") and run them directly or chain them into workflows.
Because One is framework-agnostic, it can be integrated with OpenAI, Anthropic, LangChain, CrewAI, the Vercel AI SDK, or custom-built architectures.
MCP Server Integration
The Model Context Protocol (MCP) has emerged as a common standard for linking AI tools with external services. One provides a hosted MCP server featuring over 47,500 actions across 250+ platforms.
Adding One as an MCP server in compatible environments like Cursor or Claude Desktop exposes all connected integrations to your agent immediately. If a required service is not in the default catalog, the Bridge tool allows users to input any REST or OpenAPI documentation URL to generate a hosted MCP server.
Security and Data Privacy
Because One manages OAuth credentials for numerous services, security is a core focus. The platform is SOC 2 compliant, with credentials encrypted both at rest and in transit. User connections are isolated to prevent cross-account access issues.
Key security policies:
- Integration credentials are deleted upon disconnecting an application.
- Account deletion removes all associated user data.
- No training of LLMs on user data is performed.
- Enterprise tiers offer SSO, SCIM, and detailed audit logs.
One Flow: Dynamic Workflows
One Flow is a dedicated runtime designed to process multi-step workflows. It supports conditional logic, parallel execution, loops, and pausing/resuming.
Crucially, agents can generate and edit these workflows dynamically using JSON format. This allows the agent to structure and execute custom pipelines on the fly based on runtime context, offering an alternative to static, pre-defined automation rules.
Use Cases
Common implementation patterns for the runtime include:
- Information Aggregation & Alerts: Compiling calendar events, checking incoming communications, and posting summaries to collaboration tools like Slack.
- Customer Support Automation: Retrieving order data from e-commerce platforms, reviewing email threads, and generating drafts for human review.
- Stripe Operations: Automating payment searches and financial report generation.
- Dashboard Operations: Wiring the MCP server into a frontend dashboard using frameworks like Next.js, leveraging webhooks and user-scoped credentials via AuthKit.
Strengths
- Extensive catalog: Over 47,500 ready-to-use actions across 250+ platforms.
- Platform agnostic: Compatible with any LLM, framework, or orchestration system.
- Developer-centric: CLI-driven setup and testing.
- Bridge utility: Easy generation of MCP servers for custom APIs.
- Generous free tier: 1M free API calls with no trial expiry.
Limitations
- No mobile application: Interface is strictly web and CLI-based.
- Overage costs: Usage-based pricing can scale quickly for high-volume agents.
- Recent rebranding: Rebranded from Pica in March 2026, meaning community templates and resources are still actively growing.
- Paid team features: Shared workspaces, log history extension, and advanced AuthKit features require a paid subscription.
One vs. The Competition
- One vs. Composio: Both provide MCP servers for agents. Composio focuses heavily on open-source, self-hosted deployment models. One focuses on a managed cloud runtime featuring AuthKit and the Bridge tool. Choose Composio for maximum infrastructure control; choose One for lower setup and maintenance overhead.
- One vs. Zapier: Zapier targets non-technical users with pre-built, event-driven workflows. One is built for developers requiring programmatic CLI/MCP access and dynamic agent execution.
- One vs. n8n: n8n is an open-source, self-hosted workflow automation platform. One is a fully managed runtime optimized specifically for AI agent auth and execution.
- One vs. Pipedream: Pipedream provides a developer-focused, event-driven workflow engine. One adds the specialized MCP server and user auth layers (AuthKit) required for AI tool-calling.
Ideal Users
One is highly suited for:
- Developers building AI agents who want to bypass writing custom OAuth integrations.
- Product teams looking to securely enable users to connect their own accounts via AuthKit.
- Teams requiring a single, unified integration layer across multiple distinct AI frameworks.
It is less suited for non-technical teams, simple static automations, or organizations with massive API volume on highly restricted budgets.
FAQ
What is WithOne (One)?
One is an integration runtime that provides AI agents with authenticated access to 250+ applications via a single CLI, dashboard, and hosted MCP server. It rebranded from Pica in March 2026.
Is One free?
Yes, the free tier includes 1 million API calls, unlimited connected apps, and full CLI/MCP access.
How does One handle custom APIs?
Through the Bridge feature, which converts any standard OpenAPI or REST documentation URL into a hosted MCP server.
Does One support self-hosting?
While the SDK, tools, and skills are open-source on GitHub, the runtime, Bridge generator, and AuthKit remain managed, proprietary services.
Does One train LLMs on user data?
No. The privacy policy states that user data is not used for LLM training.
How fast is the setup process?
By installing the CLI and running one connect, developers can establish their first authenticated API connection in a few minutes.
The Bottom Line
One addresses a key developer pain point in agent development: API and authentication logistics. By centralizing OAuth management, token refreshes, and tool schemas, it allows developers to quickly give their agents actionable access to external services. With a developer-friendly CLI, a large MCP catalog, and a practical free tier, it is a strong option for teams building integration-heavy AI agents.