- Part of Anywhere: Activation .
- The Fullstory for Customer Agents MCP server relies on Activation API endpoints, which consume Activations.
- 1000 Activations per month available with Enterprise and Advanced plans.
- Additional Activations available with Anywhere: Activation.
- Authenticates via OAuth 2.1—on first connection, your MCP client opens a browser window to sign in to Fullstory.
The Fullstory for Customer Agents MCP server gives AI agents real user context, on demand. Fullstory captures the complete digital journey across web and mobile—every click, navigation, error, frustration signal, and custom event. This MCP server surfaces that journey as a single summary or as structured context, optimized for LLM grounding.
This article covers the following:
Common use cases
The Fullstory for Customer Agents MCP server is designed to give AI-powered software tools—AI support agents, sales copilots, CX platforms—real-time behavioral context so they can act on what a user actually did, not just what they reported.
- Pre-call session briefings: Brief contact-center and concierge agents with what the user did before they reached out—every click, navigation, error, and frustration signal.
- Real-time grounding for AI agents: Give AI chat, voice, and copilot agents live behavioral context so they answer the user's actual question, not a generic one.
- High-intent session triggers for sales: Detect intent signals in a live session and surface enriched context to the sales play that fires next.
- Auto-enrichment of IT and CX tickets: Attach a session summary to every ticket so the resolver sees what the user actually did, not just what they typed.
Getting started
Add the server to your AI client's MCP configuration:
{
"mcpServers": {
"fullstory-customer-agents": {
"type": "http",
"url": "https://api.fullstory.com/mcp/activations"
}
}
}
Paste this block into your client's MCP configuration file, adjusting the outer structure if your client uses a slightly different schema (for example, "servers" instead of "mcpServers"). On first connection, your client will open a browser window to authorize with Fullstory.
Once connected, you can ask the agent questions like:
- "Summarize what this user did before opening the support ticket."
- "Give me the recent events and device context for this session."
- "What was the user trying to do in the last 5 minutes?"
Authentication
The server uses OAuth 2.1 with PKCE. On first connection, your MCP client opens a browser window to sign in to Fullstory and grant the required scopes. The client stores the resulting access token and refreshes it automatically.
No API keys, secrets, or service accounts are required for the standard interactive flow. If your client needs to run unattended (server-to-server), consult your client's documentation for headless OAuth handling.
Single org per connection: Each MCP server connection is scoped to a single Fullstory org. If you manage multiple orgs, create a separate connection for each.
Tools reference
All tools are available at api.fullstory.com/mcp/activations. Each tool call consumes Activations from your account quota.
generate_session_summary
Returns an AI-generated summary of a single user session, shaped by a prompt profile—a reusable configuration that controls LLM instructions, session-slicing rules, which context blocks to include, event filters, and an optional JSON response schema. Use this tool when the agent needs a narrative recap of what the user did.
Backed by the Generate Summary Server API, with optional per-call overrides matching the POST Generate Summary With Overrides endpoint. See the AI Session Summary API guide for end-to-end examples.
Parameters
| Name | Type | Required | Description |
|---|---|---|---|
session_id |
string | Yes | The Fullstory session identifier, formatted as sessionId:deviceId. If you only have a uid or email, use the List Sessions API to look up the session first. |
config_profile |
string | No | The ID of a prompt profile created via the Create Profile API. The profile defines the LLM instructions and how session data is shaped. If omitted, the default Fullstory prompt is used. |
llm |
object | No | Per-call override of the profile's LLM block. Supports pre_prompt, post_prompt, response_schema, and temperature. Non-array fields replace profile values. |
slice |
object | No | Per-call override of which portion of the session to send to the LLM. Supports mode (FIRST | LAST | TIMESTAMP), event_limit, duration_limit_ms, and start_timestamp. |
context |
object | No | Per-call override of which high-level context elements (user, org, location, device) to include. Supports include and exclude string arrays, or boolean flags like exclude_user_context. |
events |
object | No | Per-call override of which events are sent to the LLM. Supports include_types, exclude_types, trim_to_last_n_selectors, and exclude_api_events. Array fields are additive relative to the profile. |
Returns
A JSON object whose shape is dictated by the profile:
- Plain-text profile:
{ "summary": "..." } - Profile with a
response_schema:{ "response": { ... }, "response_schema": { ... } }—the response object conforms to the schema you defined.
generate_context
Returns the session's events and context, lightly transformed and structured for optimal LLM consumption. Unlike generate_session_summary, this tool does not invoke a language model—it returns the structured input your own model will consume. Use this when you want to ground your own LLM rather than have Fullstory generate the summary.
Backed by the Generate Context Server API.
Parameters
| Name | Type | Required | Description |
|---|---|---|---|
session_id |
string | Yes | The Fullstory session identifier, formatted as sessionId:deviceId. |
Returns
A JSON object containing the session information and events, formatted for input into a generative AI model. The response shape is governed by the Context specified to optimize the information for the intended use case. Refer to the Generate Context endpoint for the current parameter list as additional configuration options are added.
Frequently Asked Questions
What is Fullstory for Customer Agents designed for?
Fullstory for Customer Agents is designed to empower AI-powered software tools—such as support bots, sales copilots, and CX platforms—with real-time behavioral context about end users. It surfaces session summaries and structured event data so these tools can resolve issues faster and personalize interactions.
For an end-to-end example of connecting Fullstory session summaries to a customer-facing AI agent, see Intercom Fin—Activation Session Summaries.
Can I use Fullstory for Customer Agents with my AI code assist tool?
Although Fullstory for Customer Agents is primarily intended for customer-facing AI agents and software platforms, you can still explore its functionality in any client that supports the Model Context Protocol, including Claude Desktop, Cursor, Claude Code, and VS Code with Copilot.
If you are looking for an MCP server tailored to analytics and natural-language questions about your Fullstory behavioral data, see Fullstory MCP.
Does this MCP respect my Fullstory privacy rules?
Yes. Existing masking settings, exclusion rules, and element blocking all carry over. The MCP server operates with the same privacy enforcement as the Fullstory UI and API—it does not bypass any rules configured in your account.
Does this MCP support multi-org or umbrella accounts?
No. Each MCP server connection is scoped to a single Fullstory org. Auth tokens are org-level; cross-org querying is not supported. If you manage multiple orgs, configure a separate connection per org.
Are there rate limits?
Yes. If you encounter throttling, space out your requests or reduce the scope of your queries. If you are consistently hitting rate limits, reach out to the support team.
My client can't connect. What should I check?
Confirm the server URL is exactly https://api.fullstory.com/mcp/activations with no trailing slash. Restart your AI client—most clients only load MCP server configurations on startup. If you haven't completed the OAuth flow yet, look for a browser prompt from your client.
What are the current limitations?
- Read-only: This server does not create or modify Fullstory configuration. It returns summaries and context only.
-
Single-session scope: Each call to
generate_session_summarysummarizes one session. Cross-session aggregation is not supported—use the Fullstory MCP for analytics across sessions.