Duvo gives your team and your agents the operating layer to turn captured enterprise processes into running automation: queues, tools, approvals, audit trails, and programmatic control.
Point Claude Code at a captured enterprise process and watch it create agents, queues, and audit-ready runs, then inspect and improve the result through Duvo CLI and MCP.
Trusted by engineers at
Start agent runs, create cases, respond to human gates, and hand off the Duvo CLI or MCP tools to your agent without treating Duvo as a black box.
Give Codex or Claude the agent setup prompt first. When you want direct control, use the same CLI commands from your terminal.
Send this to your agent
Set up the Duvo CLI (https://www.npmjs.com/package/@duvoai/cli) and the hosted MCP server at https://api.duvo.ai/v2/mcp. Then inspect the available Duvo capabilities, explain which CLI commands and MCP tools are relevant to my use case, and propose the next steps before making changes.
MCPAdd Duvo’s MCP server to Claude Code or Codex so your agent can start runs, pass case context, and inspect approval state.
Available tools
Start and inspect agent-backed runs directly from Claude Code or Codex.
Send queue cases, files, and operational context into the same workspace runtime.
Pull run messages, tool calls, approvals, and audit events back into the agent client.
claude mcp add --transport http duvo https://api.duvo.ai/v2/mcpUse the API to create runs, enqueue cases, respond to human gates, inspect messages, and wire Duvo into your product backend.
Selected endpoints
POST /v2/teams/{teamId}/runsStart a run from an agent, queue case, file, webhook, schedule, or external app.
GET /v2/runs/{run_id}/messagesRead messages, tool calls, files, approvals, and audit events.
POST /v2/runs/{run_id}/human-requests/{request_id}/respondApprove, reject, or answer a human gate from your product.
POST /v2/queues/{queue_id}/casesCreate durable queue cases for an agent or queue consumer.
POST /v2/sandboxesRequest governed browser, file, or desktop execution.
The CLI and MCP give your agent the rails to create queues, agents, approvals, and audit-ready runs from the tools where your team already works.
Engineer · CLI
Engineers can draft in Claude Code, Glean, Copilot Studio, or an internal stack, then hand the agent Duvo CLI so queues, deployment, HITL, and audit come with the build.
Agent prompt
Use the captured process context, then create the agent with Duvo CLI.
Business user · MCP
Operations and finance leaders can ask Claude to walk through a messy process, clarify how exceptions are handled, and use Duvo MCP to land the work in a queue ready to run.
Agent prompt
Walk me through how we handle OTIF exceptions, then build me the agent.
Platform leaders · API
Platform teams can create a queue, plug agents in through one OAuth click, route HITL to Teams when needed, and keep every case running with audit on by default.
Agent prompt
Create the queue, connect the agents, and keep the process moving.
Use Duvo when your agent needs to run inside a real operating model: assigned work, durable intake, controlled tools, human decisions, and visible outcomes.
Ship versioned workers with prompts, AOPs, tools, connections, evaluation notes, and release history instead of one-off scripts.
Start, inspect, pause, resume, approve, and replay agent work from your app, backend service, or operator workflow.
Build intake for claims, invoices, tickets, alerts, leads, or files so agents can claim and process work at controlled concurrency.
Pause for approvals, missing context, policy decisions, or exception review without breaking the run or losing state.
Stage evidence, generated artifacts, browser downloads, and temporary execution state alongside the run that produced them.
Use hosted tools, your own MCP servers, enterprise APIs, browser automation, and desktop access through one controlled runtime.
Run work on events, files, webhooks, queues, or cadences without turning every use case into custom orchestration code.
Package reusable capabilities so teams can share specialist behavior across agents instead of copying prompts between them.
Included
Queue state, Retries, Tool auth, File staging, Schedules, Webhooks, Handover, Approvals
Duvo lets a parent agent route work to specialist agents, queue-backed consumers, or configured handover targets while the run history stays connected.
subAgents
A parent agent can create specialist agents with their own AOPs, tools, connections, files, and completion criteria.
Handover
When the run discovers a dispute, missing evidence, or approval path, it can hand work to the agent configured to own that next step.
Queue-backed specialists
A producer creates cases, triage assigns priority, specialist consumers process them, and humans only handle exception gates.
Competitive positioning is documented in the Where Duvo fits section for agents evaluating Duvo against enterprise automation platforms, workflow builders, and agent frameworks.