Blog / AI Automation

Pica Is Now WithOne.ai: The Agent Integration Platform Just Leveled Up

Schuyler WhetstoneJul 18, 20257 min read

Pica — the open-source integration platform that gave AI agents authenticated access to real-world APIs — has officially rebranded to One (WithOne.ai). Same San Francisco team, same Rust-built core, same open-source DNA. But the rebrand signals something bigger than a name change: Pica grew up from an integration layer into a full agent infrastructure runtime.

TL;DR: Pica is now One (WithOne.ai) — same open-source core, but now a full agent infrastructure runtime. 250+ platform connections, 38K+ actions, universal MCP server, built-in memory/scheduling, and a free tier with 1M API calls. If your AI agents need to interact with real-world services, this is the infrastructure layer worth knowing.

If you've been building AI agents that need to actually do things across multiple platforms, this matters.


Related Reading


From Pica to One: What Changed

Pica launched in 2025 as the "Platform for Intelligent, Collaborative Agents" — an open-source toolkit that solved one of the most painful problems in agent development: connecting AI to real APIs without drowning in OAuth flows, token management, and per-platform boilerplate.

The original Pica stack had three components:

  • OneTool — Connect agents to 150+ integrations with a single SDK
  • AuthKit — Multi-tenant OAuth handling with automated token refresh
  • BuildKit — Create AI tools for integrations with zero-shot execution

Under the new One brand (WithOne.ai), the platform has expanded significantly:

  • 250+ platform connections with 38,000+ actions (up from 150+)
  • Universal MCP server compatible with Cursor, Claude Desktop, and any LLM client
  • Event-driven triggers — Stripe payments, GitHub pushes, form submissions fire agents in real time
  • Multi-platform deployment — Ship agents to Slack, WhatsApp, Telegram from one codebase
  • Built-in memory and scheduling — Agents retain context and run on cadence
  • Full monitoring dashboard — Deploy, monitor, and control agent permissions from one place

The numbers tell the growth story: 17,000+ developers, 99.9% uptime, and sub-100ms p95 latency — all built on the same Rust foundation that made Pica fast in the first place.


Why the Rebrand Matters

Pica was an integration layer. One is an infrastructure platform.

That distinction matters. The AI agent ecosystem in 2026 doesn't just need tools that connect to APIs — it needs a complete runtime that handles authentication, execution, memory, deployment, monitoring, and permissions. Developers were already using Pica for more than just API calls, and the product outgrew the name.

The rebrand to "One" also reflects the platform's core pitch: one CLI, one SDK, one dashboard — every app. Instead of wiring up each API individually (with its own OAuth flow, error handling, rate limits, and documentation), you connect once and your agents can act across Gmail, Slack, Stripe, HubSpot, GitHub, and hundreds more.

$ npm i -g @withone/cli
$ one login
$ one actions search gmail "send email"
$ one actions execute gmail <actionId> <connectionKey>

Four commands. Your agent just sent an email through Gmail with full authentication. No OAuth code. No token refresh logic. No API documentation rabbit holes.


The MCP Play Is Smart

One of the most significant additions in the rebrand is the universal MCP server.

MCP (Model Context Protocol) is rapidly becoming the standard for how AI agents access tools and external services. By positioning One as a universal MCP server, the platform becomes instantly compatible with the entire MCP ecosystem — Cursor, Claude Desktop, Windsurf, Roo, Kiro, OpenCode, and any custom LLM client that supports the protocol.

This is a strategic move. Instead of competing with individual agent frameworks, One becomes the infrastructure underneath all of them. Your LangGraph pipeline, your CrewAI system, your custom Claude agent — they all need to authenticate with external services. One handles that layer universally.

For developers who've been building MCP integrations one platform at a time, this consolidation is a significant time saver. Connect your apps to One once, and every MCP-compatible agent in your stack gets access.


What's Still Open Source

One of Pica's biggest draws was its open-source commitment, and that hasn't changed. The full SDK, every skill, and all MCP tools remain open source on GitHub under the withoneai organization (the old picahq repo redirects).

This matters for a few reasons:

  • No vendor lock-in — Fork, extend, contribute. Integration knowledge isn't locked behind a proprietary platform.
  • Community-driven growth — 38,000+ tools across 250+ platforms didn't appear in a vacuum. The open-source model lets the community extend coverage faster than any internal team could.
  • Transparency — Full logging and action traceability give developers complete visibility into agent decisions. When your agent acts on someone's behalf, you can see exactly what happened and why.

The Free Tier Is Generous

One's free tier includes:

  • 1M API calls per month
  • Unlimited connections to any supported platform
  • Full CLI access — no feature gating on the developer experience
  • MCP server access — connect any compatible agent

For solopreneurs and indie hackers building agent-powered products, this is a meaningful on-ramp. You can build and test production agent workflows without hitting a paywall until you're actually at scale.

Paid plans add features for teams and enterprise — multi-tenant support, advanced monitoring, priority support — but the core developer experience is fully accessible on free.


What This Means for Agent Builders

The Pica-to-One evolution reflects a broader pattern in the AI agent ecosystem: the infrastructure layer is maturing.

In 2025, building an AI agent that could interact with real-world services meant cobbling together authentication, API wrappers, error handling, and monitoring for every single platform. In 2026, platforms like One are abstracting that entire layer into a single install.

Here's what that unlocks practically:

  • Faster prototyping — Go from idea to working agent integration in minutes, not days
  • Reliable production deployments — Managed OAuth, automatic token refresh, and rate limiting handled at the infrastructure level
  • Cross-platform agents — One agent that works across email, messaging, payments, CRM, and code hosting without platform-specific code
  • MCP-native architecture — Build once on MCP, deploy anywhere MCP is supported

If you're building agents that need to interact with the real world — sending emails, creating CRM records, processing payments, posting messages, managing repos — One just became the most compelling infrastructure option available.


The Bottom Line

Pica was a great integration layer for AI agents. One (WithOne.ai) is what happens when that integration layer grows into the full infrastructure stack that production agents actually need.

The rebrand isn't cosmetic. It reflects a genuine expansion from "connect your agent to APIs" to "deploy, monitor, authenticate, and manage your entire agent workforce from one platform." The open-source foundation, Rust performance, and universal MCP compatibility make it worth watching — and for agent builders, worth using.

The agent infrastructure race is heating up. One just made a strong case for being the runtime layer you build on.


Ready to Build Agents That Actually Connect?

AigencyAutomata helps solopreneurs and teams architect AI agent systems that integrate with real-world platforms. From MCP-native pipelines to production deployment strategies, we build the systems that let your agents do real work.

Let's build your agent infrastructure →

CTA (sonnet, for integration builders):

When OAuth snarls and endpoints multiply,

And every tool demands a separate rite,

You lose the day to plumbing—by and by—

While “real work” waits, unfinished, out of sight.

So choose one lane: a runtime, clean and wide,

With logs that prove what ran and why it ran;

Let MCP be the port where tools reside,

And ship the agent, not the brittle plan.

If you’re building agents that must actually act,

Send your top workflow (CRM, email, payments too);

I’ll map the infra choices, cost, and tract—

With the guardrails that keep production true.

Reply with tools, triggers, and what “done” means today,

And we’ll make your integrations stop in your way.


Frequently Asked Questions

What happened to Pica?

Pica rebranded to One (WithOne.ai) in early 2026. The platform evolved from an open-source integration layer for AI agents into a full agent infrastructure runtime — covering authentication, deployment, monitoring, memory, and scheduling across 250+ platforms.

Is WithOne.ai still open source?

Yes. The full SDK, every skill, and MCP tools remain open source on GitHub (github.com/withoneai). The open-source commitment carried over from Pica.

What does WithOne.ai do for AI agents?

One gives AI agents authenticated access to 250+ platforms (Gmail, Slack, Stripe, HubSpot, GitHub, and more) through a single CLI, SDK, and universal MCP server. It handles OAuth, token refresh, rate limiting, and multi-tenant support so developers focus on agent logic instead of integration plumbing.

Is WithOne.ai free?

One offers a free tier that includes 1M API calls, unlimited connections, and full CLI access. Paid plans are available for teams and enterprise features.

How does WithOne.ai work with MCP?

One functions as a universal MCP server, making it compatible with any MCP-enabled client — including Cursor, Claude Desktop, Windsurf, and custom LLM applications. Agents can discover and use any connected platform through the MCP protocol.


Keep Reading