MCP App Store
by KeenEthics
Indus Labs icon

Indus Labs MCP Server for ChatGPT

By Indus Labs

Overview

Indus Labs runs AI voice agents that answer and make phone calls in Indian languages and English. This app lets you build and improve those agents from ChatGPT. Describe the agent you want and ChatGPT creates it: it asks what it needs to know, picks a voice, language and models from the options in your account, writes the prompt, validates the configuration and gives you a link to talk to the agent in your browser. To improve an agent, ask ChatGPT how its calls are going: it can summarise calls over a period (volumes, durations, most common outcomes), read recent transcripts, find where callers got stuck and fix the prompt. Every change goes into a draft. Your live agent keeps running unchanged. Publishing takes two steps. ChatGPT prepares the publish, shows you what changed and any warnings, and publishes only after you approve. The confirmation expires after 10 minutes and is tied to the exact draft you reviewed. Your agent always keeps a published version, and you can undo the last publish the same way. Who it's for: businesses that run voice agents on Indus Labs for customer support, sales, order tracking and lead qualification. Limitations: ChatGPT only sees agents and calls in your own Indus Labs workspace. Phone numbers and email addresses in transcripts are masked. The app can't delete agents, buy phone numbers, start campaigns, place real calls or change billing. Indus Labs is not intended for healthcare or medical services. You need an Indus Labs account to connect, and you can revoke access at any time in your Indus Labs dashboard.

How to connect

  1. ChatGPT: Open the Indus Labs listing in the ChatGPT app directory and choose Connect. ChatGPT listing Indus Labs exposes 40 tools here.

Tools

add_knowledge_content

ChatGPT
Add a file written as text (Markdown/TXT documents; CSV for an FAQ sheet with Question and Answer columns, a product catalog or an information sheet) and build a new version. The new version is NOT live. Check it with get_knowledge_base: a sheet may wait as 'awaiting_mapping' for approve_knowledge_columns. Then publish_knowledge_version with the user's approval.

apply_improvement_edits

ChatGPT
Put chosen edits from a finished improvement run into a NEW draft (built from the version the run analysed). The live agent is unchanged; review the draft with diff_config_versions, test it, then publish with prepare_publish/publish. Ask the user which edits to apply first.

approve_knowledge_columns

ChatGPT
Approve a sheet version's column settings, with any changes (a column's meaning for catalogs, its description for the agent, or hiding it, e.g. cost price), so the version builds. Show the user the columns from get_knowledge_columns and confirm first. Only for a version with status 'awaiting_mapping'; to change a built or live sheet, upload it again with add_knowledge_content. A changed header_row_index asks for the columns to be checked again. The built version is NOT live until publish_knowledge_version.

create_agent

ChatGPT
Create a new voice agent. config.system_prompt and config.starting_instructions are required; other fields fall back to platform defaults. The first version is published immediately, so confirm the name and purpose with the user first. Returns agent_id and a dashboard_url. Read the indus://guide/agent-config resource for tips.

create_config_draft

ChatGPT
Create a new DRAFT config version for an existing agent. Only include the fields you want to change; everything else is copied from the live config. Drafts do not affect live calls until published.

create_knowledge_base

ChatGPT
Create an empty knowledge base. docs = documents & FAQs; catalog = products callers can ask about or order; table = an information sheet (dealers, stores, contacts). No agent uses it until set_knowledge_base_agents.

diff_config_versions

ChatGPT
Field-by-field diff (before/after) between two config versions of an agent.

edit_knowledge_faqs

ChatGPT
Add, change or remove FAQs of a documents knowledge base. Builds a NEW version from base_version (which isn't changed); it is not live until publish_knowledge_version.

enhance_prompt

ChatGPT
Get AI suggestions to improve a version's prompt. Nothing is saved: apply the suggestions you and the user like with update_config_draft.

get_agent

ChatGPT
Get one agent's summary, its current live config and its dashboard_url.

get_agent_insights

ChatGPT
See how an agent's calls went in a period: results against its goal, and what the analysed calls show (blockers, complaints, sentiment, recurring issues, themes, trend). Use it to decide what to improve. Contact details are removed.

get_call_analysis

ChatGPT
Get the stored analysis of one call: summary, score, issues and where the agent went wrong. Returns status 'not_analysed' if the call has no analysis.

get_call_stats

ChatGPT
Summarise an agent's calls over a period: number of calls by direction and status, total/average/longest duration, and the most common values of each extracted outcome. Counts up to 5000 calls and says when the result was truncated.

get_config_version

ChatGPT
Get the full config of one version of an agent.

get_improvement_run

ChatGPT
Get an improvement run's status and, when ready, the problems found and the edits it offers (status 'passed' or 'closest'), each with why and example evidence.

get_knowledge_base

ChatGPT
Get a knowledge base with its files and versions (status, stats, problems found when building). A version must be 'ready' before it can be put live.

get_knowledge_columns

ChatGPT
Get a catalog's or information sheet's column settings for a version: each column's header, meaning (role), description for the agent, whether it's hidden, and sample values. A new sheet's version waits as 'awaiting_mapping' until they're approved with approve_knowledge_columns.

get_knowledge_content

ChatGPT
Read what a version holds (default: the live one): document sections and FAQs (with faq_id, for edit_knowledge_faqs), or spreadsheet rows.

get_outcome_suggestions

ChatGPT
Get suggested outcome fields and how well each was extracted on recent calls (examples say which calls were checked; extracted values aren't returned).

get_recent_transcripts

ChatGPT
Read the full transcripts (with summary and outcomes) of an agent's latest calls in one step, optionally filtered by date range, direction and status. Use it to review how the agent handles real callers. Phone numbers and emails are masked.

get_test_results

ChatGPT
Get a test's status and, when ready, its scorecard: summary, calls passed, and each call's result and checks.

get_transcript

ChatGPT
Get a call's summary, outcomes and transcript turns (phones/emails masked). Use it to find where the agent went wrong, then improve the config with a new draft.

list_agents

ChatGPT
List the user's voice agents (id, name, type, active flag, live version).

list_config_versions

ChatGPT
List an agent's config versions with status (draft/published/archived) and which one is current (live).

list_improvement_options

ChatGPT
List what a voice agent can be improved on, from its real calls of the last 7, 30 or 90 days: outcome goals, common problems, themes and mistakes. Use one as the goal of start_improvement_run.

list_knowledge_bases

ChatGPT
List the user's knowledge bases (documents & FAQs, product catalogs, information sheets): type, when the agent should use it, which agents use it, live version and latest build status. With agent_id, that agent's own knowledge bases instead.

list_options

ChatGPT
List everything a config can reference: voices (voice_id, language, gender), languages, LLM/STT/realtime providers, VAD presets, guardrail groups/choices and agent types. Call this before writing tts_config, llm_config, stt_config or guardrail_choices so you only use valid ids.

list_recent_calls

ChatGPT
List an agent's calls, newest first (time, duration, direction, status, summary, extracted outcomes), optionally filtered by date range, direction and status. Pass next_before back as before to page through older calls. Phone numbers and emails are masked.

prepare_publish

ChatGPT
Step 1 of 2 to make a version live. Returns a short-lived confirmation_id plus the changes vs. the live version, warnings and the number of live sessions. Nothing goes live. Show the changes and warnings to the user and ask for explicit approval before calling publish.

prepare_revert_last_publish

ChatGPT
Undo helper: prepares re-publishing the version that was live before the last publish made through MCP. Same output as prepare_publish; still needs user approval and a publish call.

publish

ChatGPT
Publish a prepared config version so it goes LIVE for real callers. ONLY call this after you have shown the user the changes and warnings returned by prepare_publish (or prepare_revert_last_publish) and the user has explicitly approved publishing in this conversation. Never publish on your own initiative. Pass the confirmation_id from that prepare call (it expires and is single-use); also pass the version, number of changes and warnings so the user can confirm them.

publish_knowledge_version

ChatGPT
Put a built ('ready') knowledge base version live: every agent using this knowledge base answers from it on its next calls. This changes what live callers hear. ONLY call it after telling the user exactly what will change and getting their explicit approval in this conversation.

render_preview

ChatGPT
Render the final system prompt and starting instructions exactly as the agent will see them, with sample metadata filled in, plus any rendering errors (e.g. missing or invalid metadata variables). Nothing is saved.

set_knowledge_base_agents

ChatGPT
Choose which of the user's agents use a knowledge base (replaces the list; [] for none). An agent searches it on calls when its knowledge_base switch is on (set through a draft). This changes what live callers hear. ONLY call it after telling the user exactly what will change and getting their explicit approval in this conversation.

start_improvement_run

ChatGPT
Start an improvement run for a voice agent: it finds the moments in real calls where the goal is missed and tests prompt edits against them. Runs in the background for several minutes; 5 runs per agent per day, one at a time. Changes nothing until you apply edits with apply_improvement_edits. Poll get_improvement_run.

suggest_call_outcomes

ChatGPT
Suggest up to 6 call outcome fields worth extracting after each call (e.g. order id, callback time), each tested on 20 recent calls; or pass draft to test one field you propose. Background job, 3 per agent per day; adds nothing to the agent. Poll get_outcome_suggestions.

test_config_version

ChatGPT
Run about 5 simulated test calls of a config version (usually a draft) and judge them, so you can check a change before publishing. The calls use that version's prompts, voice, guardrails and end-call message; the agent's tools are simulated and an AI plays the callers. No real calls are placed and nothing is saved to the agent. Limited to 10 tests per agent per day. Returns a test_id; poll get_test_results (takes ~2-3 min).

update_config_draft

ChatGPT
Update fields of an existing DRAFT version in place; the fields you send overwrite the draft's current values. Live calls are unaffected. Fails with not_a_draft for published or archived versions; create a new draft instead.

validate_config

ChatGPT
Validate a config version. Returns valid, errors and warnings. Fix all errors before prepare_publish; mention warnings to the user.

Capabilities

Writes

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/indus-labs"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/indus-labs)