Statsig October 2026 Updates: MCP v3, Warehouse Exports, and Permissions

In October 2026, Statsig released five updates covering AI-agent integration, user permissions, analytics data exports, and experiment program management.

The central release, Statsig MCP v3, gives AI assistants access to 226 supported Console API operations. At the same time, the company expanded entity creation permissions, added error monitoring for Data Warehouse exports, introduced change previews before approval requests, and enabled dashboard filtering by experiment status.

The updates are primarily aimed at Product Analytics, Experimentation, CRO, and Software Development teams that manage numerous experiments, Feature Flags, and product metrics.

Statsig MCP v3: Expanded Platform Access Through AI Agents

On October 1, 2026, Statsig introduced MCP v3, a new version of its Model Context Protocol integration. On October 7, the company published a detailed explanation of the updated architecture.

MCP makes it possible to connect Statsig to compatible AI tools and work with experiments, Feature Gates, metrics, and configurations directly from an AI assistant.

For example, a developer can ask an AI assistant: “Show the active experiments in our project and the results of the new checkout experiment.”

The assistant retrieves existing experiments and measurement results from Statsig through MCP. Operations that modify configurations remain subject to user permissions and confirmation procedures.

From 93 Tools to 18 Core Tools

The previous Statsig MCP architecture used 93 separate tools. Each carried its own description, which was loaded into an AI agent's context. As the platform expanded, this increased the amount of information the agent needed and made it harder to select the correct operation.

MCP v3 reduces the default set to 18 tools in read-write sessions. Read-only sessions use a smaller set of 10 tools.

Core operations are available directly:

• •Reading and searching Feature Gates.
• •Working with experiments.
• •Reading Dynamic Configs.
• •Retrieving metrics.
• •Accessing event and change logs.
• •Searching project entities.
• •Determining the current context and permissions.

Additional operations are discovered only when the AI agent needs them.

In Statsig's internal tests, the new architecture reduced input-token consumption by approximately 43% for a comparable set of tasks. This is a vendor-reported internal result, not an independent benchmark of AI-agent performance.

Access to 226 Console API Operations

The main functional change in MCP v3 is broader Console API coverage. Statsig reports support for all 226 operations in the managed Console API scope at the time of release.

Instead of creating a separate MCP tool for each API method, the system uses three general-purpose mechanisms:

• •api_read: performs read operations.
• •api_write: modifies data when the required permissions are present.
• •api_destructive: handles sensitive operations that require additional confirmation.

The agent first discovers the relevant operation and its schema, then invokes it through the appropriate tool.

For example, it can retrieve an existing experiment, find an associated configuration, or prepare a change to a Feature Gate.

Importantly, api_destructive is not limited to deleting objects. Certain configuration changes and approval-related operations can also be considered sensitive. Access to each action depends on user permissions and project rules.

Structured Responses for Automation

Statsig MCP v3 returns structured data that AI agents and automated systems can process downstream. Results can therefore feed subsequent workflow steps rather than serving only as text responses.

For example, AI can retrieve an experiment, extract its status and key metrics, and prepare a team summary.

Handling of some common request-argument errors has also improved. The system can normalize safe format deviations, such as a numeric value passed as a string or differences in enum-value capitalization.

However, ambiguous identifiers and invalid parameters in potentially dangerous actions are still rejected. The goal is to make the integration more reliable, not to weaken operation validation.

Managing AI-Agent and User Permissions

On October 6, 2026, Statsig expanded Entity-Level Permissions to include permissions for creating new entities.

The company already supported separate edit and delete permissions for different entity types. Creation permissions now receive the same level of control.

For example, a company can allow product analysts to view experiments while restricting who can create new Feature Gates or Dynamic Configs. Another group could create experiments but be unable to change certain production configurations.

Entity-Level Creation Rights

Statsig supports differentiated access for these entity types:

• •Feature Gates.
• •Experiments.
• •Dynamic Configs.
• •Segments.
• •Layers.
• •Holdouts.
• •Autotunes.
• •Parameter Stores.
• •Prompts.

The actual permissions are determined by a user's role and project settings. This allows engineering, analytics, and marketing teams to divide responsibilities.

For example, a marketer may have access to experiment results and permission to create approved tests, while operations involving critical Feature Gates remain with the engineering team.

How Permissions Work Through MCP

MCP v3 uses Statsig's existing permission model. For personal keys, effective access is governed by:

• •Permissions associated with the user's role.
• •The project's maximum permitted access level.
• •Restrictions applied to the token itself.

With shared Omni or Legacy Console API Keys, individual user permissions do not apply because these keys are not tied to a specific person.

Permissions are checked at two stages:

1. When the AI agent requests the list of available operations.
2. When the agent attempts to execute a selected action.

Operations that a user cannot perform should not be offered as available tools. Teams can therefore connect AI to an active Statsig project without giving everyone the same ability to modify production configurations.

Permission to perform an operation is not the same as approval to deploy a change. An existing Review Workflow must still be followed, even when the user has the necessary technical access.

Warehouse Export Observability: Monitoring Analytics Data Exports

On October 8, 2026, Statsig introduced Warehouse Export Observability to help teams diagnose data exports from Statsig into external warehouses.

Statsig can export experiment and user-event data to corporate Data Warehouses for further analysis. Supported destinations include BigQuery, Snowflake, Redshift, Databricks, and S3.

For example, a company can export experiment exposure data to BigQuery and combine it with order information, revenue, and other business metrics.

Exports can fail because of connection errors, missing permissions, or table schema changes. If a team does not detect a failure, subsequent reporting may rely on incomplete data.

Failure Notifications and Error Details

The update introduces several ways to detect problems:

• •Email Notifications: alerts about failed exports.
• •Console Error Messages: information about the cause of a failure in Statsig's interface.
• •Console API: programmatic access to export status.
• •MCP: investigation of failures through a connected AI assistant.

For example, an analyst might ask: “Check whether any experiment data exports to our warehouse failed in the past two weeks.”

The assistant can retrieve available status details and help identify the dates and causes of unsuccessful runs. This reduces the need to inspect every export manually or contact support for initial troubleshooting.

New API for Checking Export Status

Statsig added the list-warehouse-export-status endpoint. It reports information about daily export runs, including:

• •Date.
• •Data type: Exposures or Events.
• •Export destination.
• •Success or Failed status.
• •Status code.
• •Exported row count, when available.
• •Diagnostic message.

API requests can cover a specified period and filter by data type and status. The maximum range per request is 90 days.

This enables automated export-quality checks and makes it possible to incorporate monitoring into Data Engineering workflows.

What a Missing Export Record Means

An important limitation is that no record for a given day does not necessarily indicate a failure. According to Statsig's documentation, the export might not have run, or no rows might have been available for export that day.

A missing successful status should therefore be evaluated against the expected schedule and the contents of the destination table.

MCP can surface information about failures, but it does not automatically repair Data Warehouse permissions or rerun exports. If the connected account lacks table access, those permissions need to be restored in the warehouse itself.

Data Warehouse Exports are an Enterprise feature of Statsig, making these updates particularly relevant to organizations with mature analytics infrastructure and separate BI systems.

Preview Changes Before Requesting a Review

On October 8, 2026, Statsig improved the review process for Feature Gate and Dynamic Config changes.

Before submitting an approval request, users can now preview a proposed change directly in the Review Modal. This helps them inspect a configuration before sending it to the responsible reviewer.

Visual and Diff Previews

Two views are available:

• •Visual View: a readable representation of the proposed configuration.
• •Diff View: a comparison of changed parameters with the previous state.

There are also two filters:

• •Changes: show only modified parameters.
• •All Rules: show the full set of configuration rules.

For example, a developer changes a Targeting Rule to make a feature available to an additional audience. Before requesting approval, they can confirm exactly which part of the rule changed and which settings remain unchanged.

This is especially important for production Feature Gates that control feature availability.

Selecting Reviewers and Slack Notifications

Statsig also improved the interface for choosing people responsible for approving changes. Users can search by team and individual employee.

The Review Workflow now also suggests enabling Slack Notifications. This can reduce approval delays and the need to repeat reviews because changes were prepared incorrectly.

The October release specifically concerns previews for Feature Gates and Dynamic Configs. It should not automatically be assumed to apply to every change involving Experiments, Layers, or other entities.

Filter Dashboards by Experiment Status

On October 9, 2026, Statsig added dashboard filtering by experiment status. The feature is designed for teams running many tests at once and tracking their results in shared analytics dashboards.

For example, a company might have a central CRO Dashboard. Over time, it accumulates both completed and active experiments. Previously, a team had to manage the dashboard contents manually to focus only on active tests.

Now users can choose Add Filter → Experiment Status.

How Status Filtering Works

The update supports:

• •Filtering experiments by status.
• •Applying the filter to standard Dashboards.
• •Applying it to Dashboards generated automatically from Tags.
• •Saving the selected filter as the default setting.
• •Sharing a dashboard link with a filter applied.

Widgets that do not match the selected status are not deleted. They stay in their original positions but appear visually muted.

This preserves the dashboard's layout while allowing teams to switch between different views of the experiment program.

For example, a CRO manager can open a dashboard showing only ongoing tests, while a product leader can use the same dashboard to review completed experiments and their results.

The filter does not modify experiments or traffic allocation. It changes only how the information is displayed.

Practical Value for CRO and E-commerce

The October Statsig updates are primarily intended to reduce the operational workload involved in managing experiments.

MCP v3 allows developers and analysts to work with experiments, Feature Gates, and metrics directly from their AI tools.

Entity-Level Create Permissions helps divide responsibilities without giving every employee excessive access.

Warehouse Export Observability simplifies monitoring the movement of data into corporate analytics infrastructure.

Preview Changes reduces the risk of submitting an incorrect configuration for approval.

Experiment Status Filters make it easier to monitor ongoing and completed tests.

For E-commerce companies, these capabilities can support experiments on product pages, search, recommendations, the cart, and Checkout.

For example, a team runs several parallel Conversion Rate tests. Developers manage Feature Gates and variations, analysts review results, and marketers monitor progress on a Dashboard.

MCP can accelerate information retrieval, permissions restrict configuration changes, and Warehouse Export monitoring helps find problems with experiment and event data transfers.

However, these features do not automatically generate CRO hypotheses or guarantee higher Conversion Rates. Their primary value is in organizing experiments, automating data access, and improving oversight of experimentation infrastructure.

Connecting Statsig MCP v3

Statsig published instructions for connecting MCP v3 to compatible AI tools.

Remote MCP server URL: https://api.statsig.com/v3/mcp

To connect:

1. Add Statsig MCP v3 in the settings of a supported AI client.
2. Authenticate using a supported Statsig method.
3. Confirm that the correct project is selected.
4. Check the available user permissions.
5. Run a test query for Feature Gates or experiments.

For example, a user can ask: “Which Statsig project am I connected to, and what permissions do I have?”

Then: “Show the project's active experiments.”

When moving from MCP v2, teams need to account for changed tool names and response formats.

The new version also provides the structured structuredContent field, which should be used when results are processed programmatically.

Statsig recommends disabling the old MCP connection while testing v3 to avoid confusion between tool sets from different versions.

MCP is optional: Statsig continues to offer its web-based Console and direct Console API.

Statsig: Feature Management, Product Analytics, and Experimentation

Statsig is a platform for managing product features, conducting A/B tests, and analyzing user behavior. It combines several areas:

• •Feature Management: controlling feature availability through Feature Gates, Dynamic Configs, and user allocation rules.
• •Experimentation: designing and running experiments, comparing product variants, and measuring their impact on chosen metrics.
• •Product Analytics: exploring events, user segments, funnels, and other product usage measures.
• •Session Replay: reviewing user sessions to understand interactions with the interface.
• •Warehouse Native Experimentation: running experiments and analytics with corporate Data Warehouse data.
• •Statsig MCP and Console API: programmatic access to experiments, configurations, and analytics for developers, internal systems, and AI agents.

The platform is used to develop web services, mobile apps, E-commerce systems, and other digital products.

The October 2026 updates expand the management capabilities for large experimentation programs. MCP v3 exposes more Statsig functionality to AI agents while retaining permissions and approval processes.

Warehouse Export Observability simplifies monitoring analytics integrations, while the new permission, preview, and dashboard controls help divide team responsibilities and track experimentation activity.

The practical value is less manual work and greater transparency in Product Experimentation and CRO operations.

Sources

1. Statsig — Product Updates: October 2026
2. Statsig — Introducing Statsig MCP v3, October 7, 2026
3. Statsig — Statsig MCP v3, October 1, 2026
4. Statsig — Entity-Level Create Permissions, October 6, 2026
5. Statsig — Warehouse Export Observability, October 8, 2026
6. Statsig — Preview Changes Before Requesting a Review, October 8, 2026
7. Statsig — Filter Dashboards by Experiment Status, October 9, 2026
8. Statsig Docs — MCP v3 Overview
9. Statsig Docs — Data Warehouse Exports
10. Statsig Docs — List Warehouse Export Status API