> ## Documentation Index
> Fetch the complete documentation index at: https://www.voiceflow.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Voiceflow MCP Server

> The server exposes tools for managing agents, querying knowledge, testing conversations, and inspecting results from an AI client such as Claude Code.

The server exposes tools for managing agents, querying knowledge, testing conversations, and inspecting results from an AI client such as [Claude Code](/docs/mcp/claude-code).

It's a hosted server at `https://mcp.voiceflow.com/mcp` that supports OAuth sign-in for supported clients. You sign in to Voiceflow once, and the assistant acts on the projects your account can access.

The [setup guide](/docs/mcp/claude-code) walks through connecting the server to Claude Code.

## What you can do

The server exposes Voiceflow's agent-management surface as tools, grouped by capability:

| Area | What the assistant can do |
| --- | --- |
| **Projects & environments** | List projects, manage environments, publish, compile, export, and handle variables and secrets |
| **Agent design** | Edit the global prompt, agent instructions, and routing; create and update playbooks and workflows; configure model and voice behavior |
| **Tools** | Create and wire functions, API tools, MCP tools, and prebuilt integrations; discover their inputs and execute them without a conversation |
| **Knowledge base** | Manage documents and data sources, and run retrieval queries to test coverage |
| **Testing** | Run live test conversations through the Dialog Manager and automated end-to-end simulations |
| **Measurement** | Fetch and search transcripts, design and run evaluations, and query analytics and usage |

## Run tools independently

Ask your AI client to discover a tool’s inputs and run it with explicit test values. Execution uses saved draft resources without compiling the agent. External actions are real: confirm the intended action before sending an email or changing a record.

[Function execution](/docs/mcp/execution/function) · [API tool execution](/docs/mcp/execution/api-tool) · [Integration execution](/docs/mcp/execution/integration) · [MCP tool execution](/docs/mcp/execution/mcp-tool)

The [execution guide](/docs/developers/tools/execute) covers REST, CLI, results, and troubleshooting.

## What you are approving

Connecting the server grants an OAuth token covering both reading and changing your workspace. This OAuth flow requires both scopes. Personal-key authentication is a separate option with its own access rules; see [Authentication](/docs/api-reference/authentication#mcp).

### [How MCP authentication works](/docs/api-reference/authentication#mcp)

  The two scopes, what each covers, and what a client has to support.

## How it works

- **Plain language, not tool names.** Once connected, you don't call tools directly. You describe what you want, for example "pull the last 50 transcripts for my support agent and find where it's failing", and the client picks the right tools.
- **Namespaced tools.** Every tool is exposed to the client as `mcp__voiceflow__*`, so it never collides with other servers you've connected.
- **Scoped to your account.** The assistant can only see and change the projects your signed-in Voiceflow account has access to.
- **Standard OAuth, no shared secret.** Authorization code with PKCE (`S256`), refresh tokens for renewal, and the token sent as a bearer header. Supported clients handle registration and token renewal.

**Tip:** 
  Not sure where to start? After connecting, ask the assistant to "list my Voiceflow projects". It's the quickest way to confirm the server is working and see what's available.

## Get started

### [Set up Voiceflow in Claude Code](/docs/mcp/claude-code)

  Connect the MCP server to Claude Code and sign in with your account.
