Gainsight CS MCP FAQ

What is the Gainsight CS MCP Server?

The Gainsight CS MCP Server connects Gainsight CS to LLMs such as ChatGPT and Claude using the Model Context Protocol. It allows you to interact with Gainsight data directly from your AI tool. With MCP, you can:

• Query Company and Relationship data
• Create and update CTAs and Success Plans
• Read and create Tasks within CTAs
• Log and update Timeline entries
• Build account briefs and review customer context
• Read Scorecards

All of this happens through natural language prompts, without needing to switch to the Gainsight UI.

What is the difference between Gainsight MCP and Gainsight Copilot?

Copilot is designed for users working inside the Gainsight CS UI, ideal for CSMs who want quick, preset answers without needing deep AI knowledge.

MCP is designed for AI-forward organizations using Claude, ChatGPT, or other compatible tools outside of Gainsight. It enables:

• Cross-system workflows
• Advanced automation
• AI agent integrations

Gainsight Recommendation

Start with Copilot if your team is newer to AI; use MCP when scaling and accelerating AI-driven value.

Is MCP limited to Claude and ChatGPT?

No. MCP works with any MCP-compatible platform, including Microsoft Copilot Studio and Sigma.

To connect the Gainsight CS MCP server with Microsoft 365 Copilot, refer to the Add an existing MCP server to your agent article in the Microsoft Copilot Studio documentation.

Can MCP pull data from multiple systems at once?

Yes. If you have multiple tools connected via MCP, you can ask questions that pull data from all of them. Claude or ChatGPT gathers information from Gainsight and other sources to build a comprehensive view. You can also explicitly prompt the LLM to go to specific systems. For example, “ask Gainsight and Staircase”

What can the Gainsight CS MCP do?

The MCP supports three tiers of functionality:

• Operational CRUD: Create, read, and update CTAs, Success Plans, and Timeline Activities.
• Flexible Querying: Query Company, Relationship MDA objects (companies, relationships, tasks, and more) with filtering, aggregation, sorting, and cross-object lookups.
• Read-only Configuration Inspection: View field schemas, picklist values, CTA type dependencies, Success Plan templates, and activity type configurations.

What can the MCP not do?

The MCP does not support: User Management (invite, deactivate, or assign roles), Integration or Connector configuration, Customer Goals, Rules Engine modifications, Custom Field or Schema creation, C360 layout management, Health Score configuration, bulk batch operations (currently one record at a time), Task CRUD (tasks are queryable only), notification management, delete operations on any object, or querying custom objects. Custom objects are not accessible through the MCP today. For these actions, the Gainsight UI is still required.

What Gainsight objects can MCP currently read and write?

The MCP currently supports the following setup:

• Read: Query Companies, Relationships, CTAs, Success Plans, Cockpit, Scorecards, and Timeline entries. Retrieve object metadata, templates, and picklist values.

• Create and Update: Create and update Timeline entries at the Company, Success Plan, CTA, and Relationship level. Create CTAs and Success Plans. Timeline entries at the Scorecard level are not supported.

• Search: Search Company and Relationship data across your instance.

Does the MCP support cross-account portfolio analysis?

Yes

Can MCP handle queries that span multiple objects?

MCP can currently query across Timeline and CTA data since both are available. For any object available through the MCP, the LLM can report, analyze, and answer queries in a straight forward manner.

Can I bulk-add Timeline activities for parent and child companies?

Yes. This is possible through MCP by referencing companies by name or GSID.

When creating or updating a Timeline entry, are all custom metadata fields on the type available?

Yes. All custom metadata fields configured on a Timeline activity type are available when creating or updating entries through the MCP.

Will Customer Goals be available for create or update via MCP?

No, Customer Goals is not supported.

Will Email Logs V2 be exposed to MCP?

No, Email Logs V2 is not exposed to MCP.

What are the key CSM use cases for MCP?

Top use cases include:

• Meeting prep and account briefs

• Account handoffs, for example, "I am taking over this account — give me my 30-day focus"

• Risk escalation identification

• CTA triage and cleanup

• Success Plan creation

• QBR deck building using Gainsight data alongside other tools in the LLM

Can MCP create a Success Plan from LLM output?

Yes. You can ask the LLM to generate a Success Plan based on account data and have it created directly in Gainsight. The feature works with a simple prompt referencing the account and plan type.

How does MCP work with Staircase AI?

If you have Gainsight CS and Staircase connected, Staircase Insight data that flows into Gainsight CS will be accessible when querying the Gainsight CS MCP. If you have both MCPs connected, you can also run a single prompt in the LLM that pulls from both Gainsight CS and Staircase simultaneously.

Has Gainsight MCP been connected to Glean?

Glean has a dedicated native Gainsight integration that builds a unified index for broad search. Combined with MCP support for write and action use cases, the two can complement each other for full coverage.

How do I enable the MCP Server?

Enabling the MCP Server requires coordination between two roles:

• Gainsight Admin: Creates an OAuth app in Gainsight, enables the PKCE toggle, and shares the Client ID and Client Secret with the ChatGPT or Claude admin.
• ChatGPT or Claude Admin: Creates a new app in the AI platform using the MCP Server URL and the OAuth credentials provided by the Gainsight Admin. The callback URL from the AI platform must be pasted back into the Gainsight OAuth configuration.

What is the MCP Server URL?

The MCP Server URL follows this pattern: https://[your-domain].gainsightcloud.com/v1/ds-mcp/mcp

Replace [your-domain] with your Gainsight instance domain. The authentication method is OAuth.

How do I set up MCP with other MCP-compatible tools (Microsoft Copilot Studio, Sigma, and others)?

The Gainsight setup is the same for all tools, create the OAuth app, retrieve the  MCP URL, and provide the client’s callback URL. Refer to your MCP client’s documentation for their specific setup steps.

• For Microsoft Copilot Studio specifically, an IT admin needs to create a connector in Copilot Studio first. Those connectors then become visible to end users in Copilot. You need to provide the Client ID, Client Secret, and authorization URLs.
  For more information, refer to the Add an existing MCP server to your agent article in the Microsoft Copilot Studio documentation.

• For other third-party clients such as Sigma, use OAuth as the authentication method with the MCP Server URL and OAuth client credentials. An existing OAuth app can be reused.

How does MCP handle permissions, field-level security, and mandatory fields?

MCP respects all user-level data permissions. Field-level permissions and bundle restrictions are enforced at the data layer, so an MCP query cannot return any field or object that the authenticated user cannot access in the Gainsight UI. 

UI-specific controls such as locked layouts or locked filters do not apply, since they are presentation settings rather than data permissions.

If a field is marked mandatory on a CTA layout, the system will enforce that the field is populated when creating through MCP. If a required field is missing, the API will return an error.

Can I control read-only versus read-write access through MCP?

Yes. The permissions are an intersection between the user’s application permissions and the permissions granted at the OAuth app level. If you create a read-only scoped OAuth app, MCP will return a 403 error when attempting writes, even if the user can write in the native UI.

Can I limit MCP access to a pilot group of users?

Not from within Gainsight directly. There is no explicit control within Gainsight to restrict MCP access to certain users. However, Claude and ChatGPT have distribution mechanisms at the admin level that can be used to control which users see the connector. Beyond that, limiting access from the Gainsight side would require manual auditing.

Can admins limit which MCP tools (for example, only Timeline, not CTAs) a user can access?

Not currently. Tool-level scoping within a read-write connection is not available yet, but it has been flagged as a highly requested feature. Gainsight has noted this as a valid and important use case to address.

Where does MCP run, and where does data flow when using it with an LLM?

The MCP server is Gainsight-hosted. It runs on Gainsight’s infrastructure and is accessed through a Streamable HTTP transport at your instance URL.

When your AI tool calls a Gainsight MCP tool, the data flows through the AI provider’s infrastructure (OpenAI for ChatGPT, Anthropic for Claude, and so on). The MCP server sends data to the LLM, and the LLM processes it within its session. Confirm your organization’s data policy with your security team before connecting. Using enterprise versions of Claude or ChatGPT (rather than personal accounts) is strongly recommended.

Is MCP covered by System and Organization Controls (SOC) reports?

MCP runs on Gainsight’s existing infrastructure and is covered under the same SOC 2 and security posture as the rest of the platform. The MCP server itself does not introduce a separate AI model, it is a structured API connector that the external LLM calls. AI governance controls (data handling, model usage) are the responsibility of the LLM provider.

How is MCP access restricted to approved systems and users?

Security is enforced through the OAuth setup, the callback URL, Client ID, and Client Secret are all required and within the admin’s control. Nobody can authenticate without those credentials being properly configured.

Custom connectors are not visible on personal Claude accounts, they only appear in enterprise or team accounts. When the official Gainsight CS connector goes live in the Claude directory, Gainsight plans to block common personal email domains (such as Gmail) from authenticating, following the same approach used for their plugins.

What authentication does MCP use?

OAuth 2.0 with Proof Key for Code Exchange (PKCE) at the user level. Each end user authenticates with their own Gainsight credentials.

Is there a human confirmation step before executing write actions?

The confirmation step is a feature of the LLM client, not the MCP server. Both Claude and ChatGPT show a confirmation popup before executing writes. Users can disable this in their LLM tool’s settings. The recommended approach is to require approval for all write actions.

Are there security risks from bulk actions through the MCP?

This is an active concern being addressed. Since the MCP honors existing CS user permissions, a user could potentially issue a bulk instruction, for example, “close all CTAs” through Claude or ChatGPT. The risk is similar to a user doing it manually, but faster and harder to detect.

What audit and logging visibility is available for MCP?

Operations are being logged internally. Currently, internal logs capture tool calls and the questions from LLMs, but not the full query by the user since Gainsight does not see that. Customer-facing audit visibility is a planned capability.

Admins can see which users have authorized the OAuth application, and when. 

There is currently no way to distinguish LLM-created entries from other API-based or automation-created entries. The API is idempotent by design. Gainsight has acknowledged this is an important question and is investigating further.

Do API calls via MCP count toward the tenant-level API limit?

Yes. MCP queries count toward the tenant API limit. 

How do I get the best results from MCP prompts?

A couple of simple practices go a long way:

• Use Thinking Models: Models with inference capabilities, such as ChatGPT with reasoning or Claude with extended thinking, produce better tool-calling decisions.

• Encode your business logic in the prompt: Add relevant context so the model understands how to interpret the request. For example: “At our company, a high-risk account is one with ARR over $500K and a health score below 50. Find all high-risk accounts.”.

What are some example prompts I can use?

Here’s one example.

Meeting Prep Brief

“I have a meeting with [Account Name] tomorrow. Query Gainsight for recent activities and PIR CTAs. Create a meeting prep brief with: Top 3 things to know going in, potential landmines to avoid, and suggested talking points.”

For more use cases, visit: https://www.gainsight.com/resource/mcp-use-cases/

OAuth Error: Redirect URL Mismatch

Ensure the callback URL(s) in your Gainsight OAuth Application exactly match the callback URL(s) provided by your AI tool. For Claude, ensure both claude.ai and claude.com callback URLs are registered.

Connector is not visible in the AI tool

Your AI tool’s admin may need to approve or publish the app or connector before end users can see it. In ChatGPT, workspace admins publish apps from Workspace Settings. In Claude, admins manage connectors from the organization’s settings.

I am getting 503 errors or authentication issues. What should I do?

The most common cause is a missing or incorrect callback URL. Make sure the callback URL from your AI tool has been added to the OAuth app in Gainsight. Also note that after adding a callback URL or enabling PKCE, changes can take up to 30 to 60 minutes to take effect. If the integration is not working, wait an hour and try again.

Can a customer use the Staircase AI MCP if they only have AI Follow-Up?

No. To use the Staircase AI MCP, customers must have Staircase AI  as a fully licensed product. AI Follow-Up alone is not sufficient.

Does Gainsight Copilot use the Staircase AI MCP for customers with only AI Followup?

No. Gainsight Copilot does not use the Staircase AI MCP for customers who only have AI Follow-Up.