Who can use this: Any team member
Available on: Gong Foundation
Ideal for: Sales manager, RevOps
An MCP connection gives a skill access to an external system, such as your company's research, content, or contract tools. When an agent runs with that skill, it can read data from the system and, in some cases, take action in it as part of the run.
How MCP connections work with skills
MCP (Model Context Protocol) is a standard way for AI to connect to external systems. In Gong, agents and the Assistant can't use a connection directly. It reaches them only through a skill, which lists the external tools the agent can use, and whose instructions say when and how to use them. For example, "Before preparing the account brief, look up the company's latest funding and acquisitions in your research tool."
The following table compares a skill with and without an MCP connection.
Skill without MCP | Skill with MCP | |
|---|---|---|
What the skill provides | Instructions only | Instructions plus tools to read from and act in an external system |
Data the agent can use | Gong data | Gong data plus data from the connected system |
Actions the agent can take | Actions in Gong | Actions in Gong plus actions in the connected system, such as updating a record |
Authentication | Not needed | Required |
An MCP connection exposes tools that an agent can use to get data from the external system or take action in it. For example, a Read tool retrieves data, and a Write tool creates, updates, or deletes it. The Tech admin chooses which tools are enabled for each MCP connection, so you can't turn individual tools on or off for a single skill.
Learn more about adding an MCP connection to a skill.
What agents can do with MCP
These examples show the kinds of work agents can do once a skill has a connection.
Use case | What the agent does |
|---|---|
Research an account before a call | Adds funding, acquisitions, and company details to a pre-call brief |
Prioritize accounts by buying intent | Adds buying signals to an account plan |
Bring in company knowledge | Finds internal docs and context related to the deal |
Share approved content | Finds on-message content to include in a follow-up |
Check contract status | Reports document status, signatures, and contract terms |
Understand product usage | Adds product usage data to a renewal or expansion review |
Learn why deals are won or lost | Adds customer win-loss feedback to a deal review |
Update records after a call | Updates contacts and deals with what was discussed |
Learn more about approved partners and custom MCP servers.
Roles and permissions
Different people set up an MCP connection, add it to a skill, and run the agents that use it, and each needs different access.
Tech admin: Sets up the connection in the Admin center by adding the MCP server, choosing how team members authenticate, and enabling the tools. Learn more about setting up an MCP client connection.
Skill builder: Adds the connection to a skill, which requires the Manage company agents, skills and automations permission.
Team members who run the agent: Don't need that permission. With a User-level connection, they connect their own account.
Authentication options
When a tech admin sets up an MCP connection, they choose how team members authenticate.
Option | How it works | What team members do |
|---|---|---|
Org-level | The whole company shares one set of credentials. | Nothing |
User-level | The tech admin authorizes the connection for the company first, and then each team member connects their own account. The agent acts with that account's permissions, so it can reach only the data that account can. | Connect their account before the skill can use the connection |
Manually triggered agents and the Assistant
When you run a manually triggered agent, including from the Assistant, the agent can use the tools of the MCP connections added to its skills. The Assistant doesn't use MCP connections on its own. Set up the connection in a skill, and add the skill to an agent. You can't call a skill with an MCP connection directly in the chat.
Write actions: Before the agent creates, updates, or deletes data in the external system, it asks you to confirm.
Authentication: If a User-level connection isn't connected yet, the agent asks you to connect it and runs without that connection's tools until you do. If an Org-level connection isn't set up, the agent asks you to contact your admin.
Credits: Running the agent doesn't use Gong credits.
Automated agents
To use an MCP connection in an automated agent, add a skill that has the connection to the agent's Generate with AI step. The agent can use the connection's tools only in that step.
Read-only: An automated agent can retrieve data from the external system but can't create, update, or delete data in it, because no one is present to confirm changes when a trigger starts a run. If the agent tries a write action, the run fails.
Authentication: Authenticate each connection before the agent runs, because a run fails if authentication is missing. After you connect, the agent uses the connection from the next time its trigger fires.
Credits: Each run uses Gong credits, with or without an MCP connection.
Learn more about adding skills to agents.
Frequently asked questions
What happens if an external tool fails?
If an external tool returns an error, times out, or reaches a rate limit, the agent continues with the other tools available to it, and its response can say which tool failed.
If a tool that changes data gives an unexpected result, correct the data directly in the external system.