Open-source MCP server
Shared marketing mix studies for analysts and agents
Connect compatible AI clients to the same studies, recipes, model runs and quality evidence used in Simba. Every tool is listed in a reference generated from the running server, with analysts in control of acceptance.
One shared study record
Analysts and agents work with the same backend services and persisted objects. Study history stays in Simba as work moves between the application and an AI client.
- Project-owned questions and attempt budgets
- Versioned recipes, input hashes and run lineage
- Shared progress, cancellation and results
Evidence before decisions
Evaluate saved evidence against a declared quality policy, compare candidates and record a recommendation. Analysts review the evidence and accept or reject candidates in the application.
- Preserved evaluations and decision rationale
- Missing evidence is explicit
- Fitted-window checks do not prove holdout performance
Discover, then act
Ask the connected backend which model families, transformations, priors and workflow operations it supports. Older deployments may leave capabilities unknown.
- Every tool and its parameters in the generated reference: github.com/getsimba-ai/simba-mcp/blob/main/docs/tools.md
- Guidance for prerequisites, retries and conflicts
- Existing model, scenario and optimizer workflows
Frequently asked questions
What is the SIMBA MCP Server?
It is an open-source integration that lets compatible AI clients call Simba tools for marketing mix modeling and shared Studies workflows.
Does it replace the Simba web platform?
It complements the application. Both interfaces use the same backend services and persisted studies, recipes, runs, evaluations and decisions.
Does upgrading MCP enable every backend feature?
The connected Simba backend determines feature availability. Use capability discovery before starting a workflow; an absent advertisement means unknown. Installing the MCP package does not upgrade your backend.
Can agents accept or promote models automatically?
Agents can record recommendations with evidence. Analyst acceptance remains in the frontend, and no tool automatically promotes a winning model. A passing quality report is not proof of business validity.
How do I get the new tools in an existing connection?
Upgrade or deploy the MCP server, then refresh or reconnect your client and start a new conversation. Existing sessions can retain an older tool catalog. The setup guide covers local clients, hosted authentication and ChatGPT developer-mode refresh.
Explore related SIMBA resources
Build a more transparent MMM workflow
Talk through your current data, modeling process, and where SIMBA can help.
Book a Strategy Call