# FutureSearch MCP Server [Health: Active]

**Category:** 🧮 Data Science Tools  
**Repository:** https://github.com/futuresearch/futuresearch-python  
**GitHub Stars:** 57  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/futuresearch-mcp-server

## Description
Give your AI a research team. Forecast, score, classify, or research every row of a dataset.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "futuresearch-mcp-server": {
    "command": "uvx",
    "args": ["futuresearch"]
  }
}
```

## Documentation

## What the FutureSearch MCP Server MCP server does

FutureSearch MCP Server MCP server gives an assistant access to FutureSearch forecasting and research operations. It can estimate the likelihood of a yes/no outcome, produce percentile ranges for a numeric value, estimate dates, compare categorical outcomes, or assign probabilities to threshold conditions. Forecasts include a rationale explaining the result.

Decision forecasting evaluates one outcome across multiple alternatives. Alternatives might represent a funding level, a policy choice, or another actor's possible action. The request can include context about the decision-maker, relevant facts, timing, resources, and previous actions. This information is useful when the answer depends on details that are not available on the web.

## How it works

Forecast inputs are organized as rows in a dataset. A row contains a question and, where needed, columns for alternatives, categories, thresholds, or conditions. The service returns structured output for each row rather than only a prose answer.

Supported forecast forms include:

- Binary probabilities from 0 to 100 for yes/no questions.
- Numeric percentiles, from p10 through p90, for continuous quantities.
- Date percentiles formatted as dates for timing questions.
- Joint probabilities for a list of categorical outcomes.
- Probabilities for ordered threshold conditions.
- Conditional results showing outcomes both when a premise is true and when it is false.

Decision forecasts use a separate operation from general forecasts. For a decision, the service forecasts the selected outcome under each provided option.

## Setup and configuration

The hosted MCP endpoint is `https://mcp.futuresearch.ai/mcp`. The README identifies connection instructions for Claude.ai, Claude Code, Gemini CLI, Codex CLI, and Cursor. The underlying Python SDK requires Python 3.12 or newer and installs from PyPI with `pip install futuresearch`, although an assistant using the hosted MCP endpoint does not need to install that package locally.

An API key is required for FutureSearch and can be obtained from the FutureSearch application. Forecast effort can be set to `LOW` or `HIGH` for supported operations and defaults to high; decision forecasts run at high effort. Current costs depend on the service's pricing.

## Tools and capabilities

The documented capabilities cover:

- Forecasting binary, numeric, date, categorical, thresholded, and conditional outcomes.
- Comparing the outcomes of several decision alternatives.
- Supplying shared context for questions about a user's own decision.
- Adding resolution criteria for questions with an external source of truth.
- Researching every row in a dataset with `agent_map()`.
- Returning rationales with forecast results.

## Limitations and notes

Numeric forecasts require an output field and units. Date forecasts require an output field. Categorical forecasts require a field containing the possible outcomes, and thresholded forecasts require a field containing ordered threshold conditions. Conditional forecasts return separate values for the condition and its negation.

The deprecated `everyrow` package should not be used for new work; the README recommends migrating to `futuresearch`. Pricing varies by effort and operation, so confirm current costs before running large datasets.

_Full upstream README: https://allmcps.com/mcp/futuresearch-mcp-server/readme_