# gongio-mcp [Health: Active]

**Category:** 👤 Customer Data Platforms  
**Repository:** https://github.com/JustinBeckwith/gongio-mcp  
**GitHub Stars:** 22  
**npm Downloads (last month):** 11490  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/gongio-mcp

## Description
MCP server for Gong.io - access calls, transcripts, and users

## Tools
Capabilities this server exposes over MCP:

- **list_calls** — List Gong calls with optional date filtering. Returns minimal call metadata (ID, title, date, duration). Use get_call_summary for details or get_call_transcript for full transcript.
- **get_call_summary** — Get an AI-generated summary of a single call including brief overview, key points, topics, action items, and detailed outline. This is the recommended way to understand a call - use get_call_transcript only if you need exact quotes.
- **get_call_transcript** — Get the raw transcript for a single call with speaker-attributed text. Only use this when you need exact quotes - prefer get_call_summary for understanding call content. Transcripts are truncated by default (10KB) to prevent context overflow - use maxLength and offset to paginate.
- **list_users** — List all Gong users in the organization. Returns name, email, title, and user IDs. Useful when you need a Gong user ID for search_calls filters like primaryUserIds, though for most "find calls by this person" cases you can pass their email directly to primaryUserEmails or participantEmails instead.
- **search_calls** — Search Gong calls with rich filters. The primary tool for narrowing down calls before drilling in with get_call_summary or get_call_transcript.

Supported filters:
- When: fromDateTime, toDateTime (ISO 8601). Always prefer a date range — unbounded queries pull every call in the workspace.
- Who hosted: primaryUserIds, primaryUserEmails, excludePrimaryUserIds.
- Who participated (host OR attendee OR invitee): participantUserIds, participantEmails, excludeParticipantUserIds, excludeParticipantEmails.
- Customer/topic: customerName (CRM account name, email domain, or title substring), titleContains, trackers (see note below).
- Metadata: scope (External/Internal), direction, system (Zoom/Meet/…), language (eng/jpn/…), minDuration and maxDuration in seconds.
- Output shape: include (array of keyPoints, trackers, highlights, speakers, comments, context, outline, media). Parties + brief + topics are always returned.

Behavior:
- Auto-paginates up to ~5000 calls. If a user asks for a broad question, guide them to narrow with a date range, scope, minDuration, or customerName first.
- trackers filter does case-insensitive substring match on tracker names. Names are workspace-specific — call get_trackers first to see what's configured before guessing.
- When the rich output would exceed the output cap, the tool auto-falls back to a compact table showing all IDs/titles. Use get_call_summary on specific IDs to go deeper.
- Filters compose with AND logic (primaryUserIds + customerName = hosted by user X on customer Y calls).

Usage pattern: narrow with search_calls → drill into specific calls with get_call_summary (AI summary) or get_call_transcript (exact quotes).
- **search_calls_by_account** — Find calls involving a specific account/company by matching email domains of external participants. The Gong API does not natively support filtering by account name — this tool fetches calls in the date range and post-filters on parties[].emailAddress. Auto-paginates up to maxCalls. For external tech-stack joins (e.g., "all calls with prospects on Klaviyo"), resolve domains upstream and pass them here.
- **search_calls_by_opportunity** — Find calls linked to specific CRM Opportunities by ID or name substring. Requires Gong-CRM integration (Salesforce/HubSpot) — calls without CRM linkage will not match. Provide opportunityIds OR opportunityNames (or both).
- **search_transcripts** — Free-text keyword search across call transcripts within a bounded date range. Use for ad-hoc searches like "calls mentioning competitor X". For recurring terms, prefer setting up Gong Trackers in the UI and using search_calls + get_call_summary — Trackers are server-side and dramatically cheaper. Date ranges > 30 days require additional narrowing via primaryUserIds or domains. Returns sentence-level matches with speaker attribution and timestamps.
- **get_call** — Get metadata for a specific Gong call including URL, direction, scope, system, and duration. Faster than get_call_summary when you only need call metadata.
- **get_trackers** — List keyword tracker definitions configured in a workspace, including every tracked phrase and which side is tracked (company/customer). Call this before using the trackers filter on search_calls so you know what names exist — workspace admins set these up and naming varies (e.g., "Competitors" vs "Competitor Mentions"). Also useful to explain tracker hits that appear in call summaries.
- **get_user** — Get a specific user profile including name, email, title, phone, and settings. Use to resolve user IDs returned from call data.
- **search_users** — Search and filter users by IDs or creation date. More flexible than list_users for resolving specific user IDs from call data.
- **list_workspaces** — List all Gong workspaces with their IDs and names. Use workspace IDs as filters in list_calls, search_calls, get_trackers, and other tools.
- **list_library_folders** — List all public Gong call library folders for a workspace. Returns folder IDs and names to use with get_library_folder_calls. Use list_workspaces to find workspace IDs.
- **get_library_folder_calls** — Get all calls saved in a specific Gong library folder. Returns call IDs, titles, curator notes, and snippet timing. Use list_library_folders to find folder IDs. Call IDs can be passed to get_call_summary or get_call_transcript.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "gongio-mcp": {
    "command": "npx",
    "args": ["-y","gongio-mcp"],
    "env": {
      "GONG_ACCESS_KEY": "",
      "GONG_ACCESS_KEY_SECRET": ""
    }
  }
}
```

**Requires environment variables:** `GONG_ACCESS_KEY`, `GONG_ACCESS_KEY_SECRET` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What gongio-mcp does

gongio-mcp MCP server connects an MCP client to Gong.io and makes selected Gong data available as agent tools. The available operations cover calls, transcripts, users, workspaces, keyword trackers, and public call library folders.

For call analysis, it can return basic metadata, AI-generated summaries, key points, topics, action items, detailed outlines, and speaker-attributed transcript text. Search functions support filters such as date range, workspace, participant, customer, account email domain, CRM opportunity, title, duration, scope, and transcript keywords.

## How it works

The server authenticates to Gong with an access key and secret supplied through environment variables. An MCP-compatible client starts the process, and the client can then invoke the exposed tools with their documented parameters.

Call retrieval is split between lightweight and detailed operations. `list_calls` returns minimal records and supports date, workspace, and cursor parameters. `get_call` retrieves metadata for one call, while `get_call_summary` and `get_call_transcript` provide progressively richer content. Transcript responses use character-based pagination through `maxLength` and `offset`.

`search_calls` automatically requests multiple API pages, up to roughly 5,000 calls or 50 pages, then applies supported client-side filters. Its default result format includes call metadata, summaries, topics, and participants.

## Setup and configuration

Requirements are Node.js 18 or newer, or Docker, plus Gong API credentials. Gong administrators can create an API key from Company Settings, Ecosystem, and API. The access key and secret must both be available when the server starts.

Installation options include running `npx gongio-mcp`, installing the package globally with npm, building from the repository, or building and running the supplied Docker image. Configure these variables:

- `GONG_ACCESS_KEY`
- `GONG_ACCESS_KEY_SECRET`

The README provides launch examples for Claude Desktop using either npx or Docker, and for Claude Code using `claude mcp add`. Repository builds use `npm install`, `npm run build`, and `node dist/index.js`.

## Tools and capabilities

The gongio-mcp MCP server includes tools for:

- Listing calls by date, workspace, or cursor.
- Retrieving call metadata, AI summaries, and paginated transcripts.
- Searching calls by structured fields, accounts, opportunities, or transcript text.
- Listing keyword trackers and workspaces.
- Listing public library folders and retrieving their saved calls.
- Getting one user, searching users, or listing workspace users.

Use `list_workspaces` when another operation requires a workspace ID. Use `get_call_summary` for an overview and `get_call_transcript` when exact spoken text is needed.

## Limitations and notes

Large searches can exceed the MCP output limit, which defaults to 50,000 characters and can be changed with `MAX_MCP_OUTPUT_LENGTH`. When that happens, the server falls back to a compact table while retaining call IDs and titles. Narrow the date range, add scope or customer filters, set a minimum duration, or remove expensive content options such as `outline`.

The `outline` include option can produce about 80 KB per call, so it is unsuitable for broad searches. Tracker output is limited to matching trackers, or trackers with nonzero counts when no tracker filter is supplied. Transcripts are paginated by default to reduce context size; request later sections with an offset.

_Full upstream README: https://allmcps.com/mcp/gongio-mcp/readme_