OneRenamed
A hosted connection layer for AI agents, formerly Pica: you connect each app once in your own account, and every agent reaches it through one MCP server or CLI, with per-action scopes enforced and logged on each call.
- What it is
- Agent tools and integrations
- Type
- Commercial product
- Made by
- One Systems, Inc.
- License
- Proprietary service; knowledge base under the One Knowledge Commons License
- Status
- Renamed. Called Pica until 25 March 2026, when the platform was rebranded as One and moved to withone.ai.
- Facts checked
- 13 October 2026
What it is
One sits between agents and the apps they act on. A person connects Gmail, Slack, Stripe or any of the 972 or so supported platforms once, in their own One account, and sets what each agent may do, down to single actions such as stripe.read without stripe.write. Agents never get the credential: One keeps it encrypted, attaches it when an action runs, checks the grant first, and declines anything outside it with a 403 that never reaches the platform. Every call is logged with the agent, app, action and result. One says it is a passthrough that stores none of the data flowing through it.
Agents reach One in two ways. The hosted MCP server at https://mcp.withone.ai/mcp (Streamable HTTP, OAuth sign-in) exposes three tools, to list integrations, find actions with their documentation, and execute them, in about 1,200 tokens of context however many apps are connected. The one CLI does the same from a terminal and installs a skill into coding agents. Behind both is a knowledge base of about 130,000 actions, each run against a sandbox before agents may use it; it is published on GitHub under One’s own Knowledge Commons license, which requires attribution, while the platform itself is a closed hosted service.
Every plan includes 1 million requests a month; past that it costs $20 per million. Free allows 20 requests a minute, Starter ($29 a month) 100 and Pro ($199 a month) 500, and paid plans mainly raise limits, seats and log retention.
Why teams pick it
- Connections belong to the user, not to each agent: connect an app once and use it from Claude Code, Cursor, Codex or any MCP client, then revoke it in one place
- Scopes are checked per action at call time and every call is logged, which gives a record for a security review
- Small context cost: three MCP tools in about 1,200 tokens regardless of how many of the 972 platforms are connected
- Generous free tier: 1 million requests a month, and adding Connect to your own product costs nothing
Watch out for
- A hosted middleman: credentials live in One’s cloud and every action passes through it, and only the knowledge base is published, under a custom license rather than an OSI one
- Low rate limits on cheaper plans: 20 requests a minute on Free and 100 on Starter, counted per account across every key
- Renamed from Pica in March 2026, so older SDKs, docs at docs.picaos.com and the GPL-licensed Pica community edition (no longer maintained) describe an earlier product
- Counts change quickly (500+ platforms and 80,000+ tools in July 2026, 972+ and 130,000+ by October), so check coverage for the apps you need
Alternatives
What all agent tools and integrations have in common →Compare side by side →
ComposioCommercialA hosted tool layer for AI agents: one session per end user gives the model a few meta tools to search, authorize and run actions across more than 1,500 app toolkits, with Composio holding and refreshing the OAuth tokens.
GlamaCommercialThe largest public directory of MCP servers, which builds and runs each open-source server in a Firecracker sandbox to record its tools, score their descriptions and flag risky behavior, plus hosting and a gateway for the servers you use.
Model Context Protocol (MCP)Open sourceAn open protocol, built on JSON-RPC, that lets an AI application discover and call tools, read resources and use prompt templates from any server that speaks it, so one integration works across Claude, ChatGPT, Cursor, VS Code and other clients.
OpenAPI SearchOpen sourceA small, open directory of OpenAPI descriptions for popular APIs, with an MCP server that turns any of them, or any spec URL, into a short plain-language summary an AI model can read before writing calls.
SpeakeasyCommercialGenerates typed SDKs, Terraform providers, CLIs and MCP servers from an OpenAPI document, and now also sells a hosted MCP platform and an enterprise MCP gateway that governs which agents may call which tools.How it fits with other technologies
Connect an app once, then find and run actions from the CLI · bash
npm install -g @withone/cli
one init # sign in, install the One skill into your agents
one add gmail # connect Gmail once, in your own account
one list # what am I connected to?
one actions find gmail "send an email" # the action and its docs
one actions execute gmail <actionId> <connectionKey> \
-d '{"to": "jane@example.com", "subject": "Hello", "body": "Sent from my AI agent"}'
# Or skip the CLI and point an MCP client at the hosted server
claude mcp add --transport http one https://mcp.withone.ai/mcpSee it in a real stack
No architecture uses One yet. Drop it into the playground and wire it up yourself.Open playground →Where it came from2021–2026
One began as Pica, an integration platform for AI agents with managed OAuth and tools for frameworks such as the Vercel AI SDK and LangChain. Its company says it raised its first seed round in 2021 and draws on 12 years of experience in integrations. Pica published a GPL-licensed community edition on GitHub in 2024, later marked as no longer maintained when Pica moved to a private, cloud-hosted platform. A v2 platform in July 2025 listed over 21,000 actions across 200 integrations.
On 25 March 2026 Pica became One, at withone.ai, rebuilt around agents rather than integrations: Auth for managed OAuth, Relay for webhooks, a CLI, and an open knowledge base of tool definitions, its GitHub repository created that month. A hosted remote MCP server followed on 13 July 2026, with consent screens that scope an agent to read-only, full access or a set of actions, enforced on the server.
In July 2026 One announced $4 million in new seed funding, $7 million in seed money in total, from investors including Four Rivers, Ripple Ventures, Golden Ventures and Village Global, and described itself as “the trust layer for AI agents”: the place where authentication, permissions and an audit record sit between an agent and a company’s software.
Why it’s built this way
- Connections owned by the user, not by the agent or the app
- One argues that the handover of credentials is where agents break: the same login pasted into every agent, never scoped or revoked. Keeping one connection per app in the user’s own account lets every agent share it and lets the user revoke it once.
- Enforce scope when the action runs
- A grant checked only at setup cannot stop a later write, so One checks every call against the grant, returns a 403 for anything outside it, and logs the attempt.
- Three MCP tools instead of one tool per action
- Tool stacks that add definitions for every integration spend the agent’s context on schemas; list, find and execute keep the footprint near 1,200 tokens at 5 platforms or 972.
- Test every action before agents use it, and publish the results
- Agents guess wrongly at parameters and response shapes, so each action is run against a sandbox (auth, endpoint, parameters, body, response, errors) and the knowledge base is public so one fix reaches everyone.
Timeline
- 2021
First seed round, per the company
- 2024
Pica community edition published on GitHub under GPL-3.0
- 2025
Platform v2 in July with over 21,000 actions across 200 integrations
- 2026
Pica becomes One on 25 March, with a CLI, Relay and an open knowledge base
- 2026
$4 million in new seed funding announced on 6 July
- 2026
Hosted remote MCP server with server-side scopes on 13 July