Integrations

Unflood works inside the AI tools your team already uses

Nobody has to learn a new app. Your people keep building and using agents in Claude, Codex, ChatGPT or Copilot. Unflood connects to each of them, so every agent is tested, approved and shared the same way, wherever it is used.

Build where you think. Use it where you work.

An agent usually starts in one person's chat. Unflood turns it into something the whole team can rely on, then delivers it to the platforms where your colleagues already are.

  1. 1 · Build

    In a chat

    Someone describes the agent in Claude or Codex, the way they already work.

    ClaudeCodex
  2. 2 · Make it reliable

    In Unflood

    Unflood tests it, versions it and asks a person to approve each release.

    TestedApproved by a personRuns in your environment
  3. 3 · Use

    Where your team works

    Colleagues use the approved agent in the tool they already have, on your company's own plan.

    ClaudeCodexChatGPTMicrosoft 365 CopilotGemini EnterpriseAzure AI Foundry

How Unflood makes it work

  • One connection, every platform

    Unflood connects to each platform through the open standard they all speak, the Model Context Protocol (MCP). The same agent, with the same tests and approvals, reaches every platform without being rebuilt.

  • Your platform does the work

    The agent runs in your team's own Claude, ChatGPT or Copilot, on the plan you already pay for. Unflood keeps the record: what the agent is, how it was tested, who approved it.

  • The same rules everywhere

    Decide once who can use, change or release an agent. The rule holds on every platform, and every change is logged.

Platform by platform

What your team does on each platform, how Unflood shows up there, and who sets it up.

Add in minutes

  • Claude

    Your team
    Builds agents in a chat, and uses the ones colleagues shared.
    Unflood shows up as
    A connector in Claude, and a plugin for Claude Code.
    Set up by
    Anyone, as a custom connector; or an Owner, once for the company.
    See how it works: Claude
  • Codex

    Your team
    Developers build and release agents from Codex.
    Unflood shows up as
    A Codex plugin on the same connection.
    Set up by
    The workspace admin, from our plugin marketplace.

Connected for you by our engineers, from day one

  • ChatGPT

    Your team
    Uses the agents your team has released.
    Unflood shows up as
    An app in ChatGPT, on the same connection.
    Set up by
    Your ChatGPT admin, with our engineers.
  • Microsoft 365 Copilot

    Your team
    Uses released agents inside Microsoft 365.
    Unflood shows up as
    An agent in Microsoft 365 Copilot, in your own tenant.
    Set up by
    Your Microsoft 365 admin, with our engineers.
  • Gemini Enterprise

    Your team
    Uses released agents from the company's Agent Gallery.
    Unflood shows up as
    An agent in your Gemini Enterprise Agent Gallery.
    Set up by
    Your Google admin, with our engineers.
  • Azure AI Foundry

    Your team
    Runs released agents in its own cloud.
    Unflood shows up as
    An agent in Azure AI Foundry, in your own Azure subscription.
    Set up by
    Your Azure team, with our engineers.

Open standards, for the frameworks your developers use

  • Agent frameworks

    Your team
    Developers build agents in code, in LangChain and LangGraph, CrewAI, the OpenAI Agents SDK, Google ADK, Microsoft Agent Framework, LlamaIndex.
    Unflood shows up as
    An MCP server their agents call for tools and approved agents, and A2A agents they can hand work to.
    Set up by
    Your developers, with Unflood's MCP address and each agent's A2A card.

Product names are trademarks of their owners and are used only to say which platforms Unflood works with. Unflood is not endorsed by or affiliated with them.

For your business, IT and legal teams

Built to pass your review

What each team checks before an agent reaches your people, and how Unflood answers it.

Business

Fits the AI tools you already pay for

Nothing to replace and no new place for your people to work.

  • Your platforms, your plans

    Agents run in the AI platforms your company uses, on the plans you already have.

  • One agent, every platform

    Each agent is recorded once, with its tests and approvals, and delivered to every platform you use.

  • Installed the way you prefer

    On Claude, an Owner adds Unflood once for the company. On the enterprise platforms, our engineers set it up with your IT team.

Business

Proven before anyone uses it

Agents reach your people the way software does: tested, reviewed, released.

  • Tests and evidence per version

    Every version carries its tests and the evidence they produced.

  • Two people, not one

    A person in your team approves a version, and a second person who did not build it promotes it to your users.

  • Agents propose, people decide

    Unflood's agents draft and prepare. Approving, promoting and changing policy always take a person.

IT and security

Your identity provider, your directory

Access follows the accounts and groups you already manage.

  • Work sign-in

    People sign in with their Microsoft or Google work account, limited to your verified domain. SAML or OIDC with your own identity provider is set up with your IT team.

  • Leavers handled

    When your directory deactivates someone (SCIM from Microsoft Entra ID), their access ends at the next call, and their agents and open approvals pass to a remaining owner.

  • Access per person

    Each person gets their own token, issued for Unflood's address only (RFC 8707). No shared credentials.

IT and security

Control over what agents can do

Every action is labelled, so your rules decide, not the prompt.

  • Read, write or irreversible

    Each tool says what it does. Irreversible actions always ask a person, and your admins can require approval or block any tool.

  • Behind your gateway

    A published allowlist names every tool and whether it reads or writes, so your MCP gateway lets through only what you choose.

  • Safe with outside content

    Once Unflood's agents read outside content, they can only read and propose. Writes follow each tool's label and your approval rules.

Legal and procurement

Your data stays yours

Unflood holds how your agents work, never the business data they work on.

  • No business data held

    Agents reach your systems through grants you consent to, and the data stays there. Unflood receives only what a platform sends in a tool call, not your conversations.

  • No model training

    Unflood does not train models. Your agents run on your AI platform, under your plan's terms.

  • No secrets in agents

    Agents and packages never contain tokens; connections hold references to the secret. Your admins can read the subprocessors that apply to your organisation at any time.

Legal and procurement

A record of everything, and a clear exit

What you need for an audit, an incident or a change of mind.

  • A tamper-evident log

    Every change, approval and stop is recorded: who, what, through which tool, and why.

  • Stop an agent everywhere

    One command pauses its triggers, revokes its access and withdraws each platform's copy.

  • Yours to keep

    Agents export as Claude Code plugins, A2A agents or git repositories. Controls and evidence are mapped to the EU AI Act, ISO 42001 and NIST; we hold no certification yet and start SOC 2 when a contract calls for it.

Read the Trust page

See it with one of your own agents

Talk to us: we show you how to connect Unflood to Claude yourself, or our engineers connect it to your company's platforms.