Anthropic

Anthropic’s Claude and Claude Code both let you run your Synthflow workspace from a conversation. Connect the Synthflow connector once, then create and manage agents, place and review calls, build knowledge bases, and run actions by describing what you want in plain language, without leaving the tool you already use. The connector works the same way in both.

This is a good fit when your work already lives in Claude.

What you can do

Ask Claude to:

• Build and update agents: Create agents, draft and refine prompts, publish drafts, and compare versions before you go live.
• Manage calls: Place test calls, review recent calls, and surface summaries or negative-sentiment moments when you ask.
• Manage knowledge: Create knowledge bases, add items from files or URLs, and attach them to the agents that should use them.
• Run actions and simulations: Scaffold and test actions, then run simulations to validate behavior before launch.

Synthflow exposes these capabilities as individual tools, grouped into read-only tools that look things up (such as listing agents or reviewing calls) and write and delete tools that change your workspace (such as creating an agent or removing a knowledge base item). Claude lets you set an approval policy for each group, so you control what runs automatically and what waits for your confirmation. Keep write and delete tools behind approval unless you are sure. For the full list of tools, see the capabilities table.

Setup instructions

The Synthflow connector works the same way in both Claude and Claude Code. You sign in through Synthflow, so there is no API key to copy or manage by hand. Know which region your workspace belongs to, then connect the Synthflow connector to the matching MCP server URL:

You are redirected to Synthflow to log in, where you select the workspace you want to manage.

After you select the workspace, click Allow access on the next screen and the connection is established. You can then start prompting.

Example prompts

Security considerations

Connecting an AI tool to operational tools introduces new risks. The primary risk class unique to these workflows is prompt injection: untrusted content (like call transcripts, user messages, or knowledge base documents) can include instructions that try to trick the model into taking unintended actions. Keep confirm before running tools enabled and review each action carefully.

FAQ