PawFlow

Integrations

Bring your model. Expose one runtime.

Mix providers, clients, protocols, and infrastructure without rebuilding your agents and durable context around each one.

Install PawFlowIntegration docs
External AI client connected to PawFlow through MCP
Other AI systems can reach the same governed tools and context.

External interfaces

MCP, ACP, A2A, and AG-UI belong to the same runtime.

PawFlow can expose its governed runtime and connect external agent processes without discarding durable state.

MCP

Tools and context

Give compatible LLM clients controlled access to PawFlow tools, prompts, and live Relay-backed workspaces.

Connect MCP
ACP

Outbound agent processes

Launch an explicitly configured ACP v1 agent, negotiate its session and capabilities, and expose only approved PawFlow MCP or client filesystem methods.

Connect ACP
A2A

Agent interoperability

Publish and discover agent capabilities for agent-to-agent collaboration across systems.

A2A reference
AG-UI

Application interaction

Embed PawFlow agents into frontends and applications while retaining runtime state and control.

Embed an agent

Works with your stack

Switch components without throwing away the runtime.

Direct APIs

OpenAI Chat Completions and Responses, Anthropic, OmniRoute, DeepSeek, and OpenAI-compatible endpoints.

Native CLIs + agents

Claude Code, Codex, Antigravity/Agy, Gemini CLI, managed native-hook variants, and installed ACP v1 agent commands.

Clients

Web, PawCode CLI, VS Code, Android, Telegram, API, and embedded AG-UI experiences.

Infrastructure

Docker, VPS, local workstation, remote machines, private services, and GPU hosts.

Secrets + identity

Local configuration or external secret providers with permissions, provider-owned refresh policy, and policy gating.

Live workspace demo

Ask ChatGPT about your real project—without uploading it.

ChatGPT → MCP → PawFlow → Relay → MyWorkspace.

No ZIPThe client sees current files through the Relay route.
No stale copyAnswers can be grounded in the live workspace.
Read-only when neededExpose observation without granting write access.
A published MCP workspace connected to an external client.

Keep the shared state

Change models, not your whole stack.

Configure providers per agent or conversation while context, tools, permissions, and clients remain in PawFlow.

Configure providers