Indus Labs MCP Server

The Indus Labs MCP server lets you build, test and improve your voice agents from the AI assistant you already use: Claude, ChatGPT, Cursor or any client that supports the Model Context Protocol. Describe the agent you want, and the assistant creates it, writes drafts, checks them, reads your call transcripts and, with your approval, publishes the changes.

Server URL

https://mcp.induslabs.io/mcp

You need an Indus Labs account. Sign up or log in at playground.induslabs.io. The assistant can only see and change agents in your own workspace.

Connect your assistant

Claude and ChatGPT sign you in with your Indus Labs account (OAuth). Developer tools such as Claude Code and Cursor use a personal access token, which you create in the MCP console.

Claude (web & desktop)
ChatGPT
Claude Code
Cursor & others

In Claude, open Settings → Connectors.
Find Indus Labs in the directory and click Connect. If you don't see it, click Add custom connector and enter https://mcp.induslabs.io/mcp.
Log in to Indus Labs and click Allow.
In a new chat, turn on the Indus Labs connector from the + menu and ask Claude to list your agents.

On Team and Enterprise plans, an Owner may need to add the connector for the organization first.

In ChatGPT, open Settings → Apps & Connectors → Advanced settings and turn on Developer mode.
Click Create, name it Indus Labs, enter https://mcp.induslabs.io/mcp and choose OAuth.
Log in to Indus Labs and click Allow.
In a new chat, pick the Indus Labs connector and ask it to list your agents.

Open the MCP console and create a personal access token. It is shown once, so copy it.
Run this command with your token:

claude mcp add --transport http indus https://mcp.induslabs.io/mcp \
  --header "Authorization: Bearer indus_pat_..."

Create a personal access token in the MCP console, then add the server to ~/.cursor/mcp.json (or your client's MCP settings):

{
  "mcpServers": {
    "indus": {
      "url": "https://mcp.induslabs.io/mcp",
      "headers": { "Authorization": "Bearer indus_pat_..." }
    }
  }
}

What you can ask

Build an agent from a description

"Create a Hindi and English voice agent for Sharma Dental that books, moves and cancels appointments. Use a warm female voice."

Claude checks the available voices and languages, creates the agent and a draft, validates it, and gives you a link to test it in your browser.

Improve an agent from real calls

"Read the last 10 calls of my collections agent and tell me where callers got confused. Then fix the prompt in a draft."

Claude reads the transcripts, summarises the problems, and writes the fixes into a new draft. Your live agent is not changed.

Publish safely, or go back

"Show me what changed in the draft and publish it if it looks right." or "Undo the last publish."

Claude shows the differences and any warnings, asks you to approve, and only then publishes. Reverting uses the same approval step.

Tools

Read tools change nothing. Write tools create agents, drafts or Gem sessions but never change a live agent. Overwrite tools replace a draft or brief in place, and Destructive means it changes what your callers hear. Your assistant asks you before either.

Tool	What it does	Type
Discover
List agent options `list_options`	Voices, languages, LLM, STT and guardrail choices you can use.	Read
List agents `list_agents`	Agents in your workspace.	Read
Get agent `get_agent`	One agent and its live configuration.	Read
List config versions `list_config_versions`	Published and draft versions of an agent.	Read
Get config version `get_config_version`	One configuration version in full.	Read
Compare config versions `diff_config_versions`	What changed between two versions.	Read
Build
Create agent `create_agent`	A new agent (it gets no phone numbers or traffic).	Write
Create config draft `create_config_draft`	A draft version; the live version is untouched.	Write
Update config draft `update_config_draft`	Change a draft: prompt, voice, language, models. Overwrites the draft's current values.	Overwrite
Validate config `validate_config`	Check a version for problems.	Read
Preview rendered prompt `render_preview`	The prompt the agent will use, with variables filled in.	Read
Suggest prompt improvements `enhance_prompt`	Suggested prompt changes for a version.	Read
Publish
Prepare publish `prepare_publish`	The changes and warnings, and a confirmation for publishing.	Read
Prepare revert of last publish `prepare_revert_last_publish`	Same, for going back to the previous version.	Read
Publish agent version `publish`	Make a prepared version live. Needs your approval.	Destructive
Gem builder
Start Gem builder `gem_start`	Start a guided agent-building session.	Write
Answer Gem question `gem_answer`	Answer the builder's next question.	Write
Update Gem brief `gem_update_brief`	Replace the agent brief with an edited copy.	Overwrite
Answer Gem build gaps `gem_gaps`	Answer the details a build is missing; starts a rebuild.	Write
Build agent with Gem `gem_build`	Generate the agent from the brief.	Write
Get Gem session status `gem_status`	Progress of a Gem session.	Read
Create agent from Gem `gem_create_agent`	Save the generated agent to your workspace.	Write
Calls
List recent calls `list_recent_calls`	Recent calls for an agent.	Read
Get call transcript `get_transcript`	A call's transcript, with phone numbers and emails masked.	Read
Get test call link `get_test_call_link`	A dashboard link to talk to the agent in your browser.	Read

How publishing stays safe

Changes go into a draft first. Your live agent keeps running unchanged until you publish.
Publishing takes two steps. The assistant prepares the publish, shows you what changed and any warnings, and publishes only after you approve.
The confirmation expires after 10 minutes and stops working if the draft changes, so you always approve exactly what goes live.
Your agent always has a published version. If something is wrong, ask the assistant to undo the last publish.
The assistant can't delete agents, buy numbers, start campaigns, place real calls or touch billing.

Permissions

When you connect, you approve these permissions for that app:

Permission	Allows
`agents:read`	See your agents, versions and options.
`agents:write`	Create agents and drafts.
`agents:publish`	Publish a version or undo a publish, after your approval.
`gem:use`	Use the Gem agent builder.
`calls:read`	Read your recent calls and transcripts.

Data and privacy

The assistant only sees agents and calls in your own workspace.
Phone numbers and email addresses in transcripts are masked before they reach the assistant.
The MCP server doesn't store your data. Requests go straight to the Indus Labs platform, which keeps an audit log of changes made through the MCP server.
Access tokens expire after an hour and are refreshed automatically. Personal access tokens expire after 90 days.
See our privacy policy and terms and conditions.

Disconnect

Open the MCP console. Under Connected apps, revoke the app or token. It stops working at once. You can also remove the connector in your assistant's settings.

Troubleshooting

"Your Indus Labs credential is invalid or expired": disconnect and connect again, or create a new personal access token.
"Rate limited": wait for the time in the message and try again. There are daily limits on creating agents, drafts and publishes.
The agent isn't found: the assistant can only use agents your account owns.

Need help? Email hello@induslabs.io. To build agents with code instead, see the Agent Management APIs.