# Everyrow MCP Server [Health: Active]

**Category:** 🧮 Data Science Tools  
**Repository:** https://github.com/futuresearch/everyrow-sdk  
**GitHub Stars:** 57  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/everyrow-mcp-server

## Description
AI-powered dataframe ops: transform, dedupe, merge, rank, and screen with natural language

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

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

## Documentation

## What Everyrow MCP Server MCP server does

Everyrow MCP Server MCP server exposes FutureSearch capabilities to assistants through the Model Context Protocol. The underlying service turns questions about future events and decisions into structured results such as probabilities, dates, numbers, or categorical outcome probabilities. It also supports research operations for building forecast inputs or answering questions that are not about the future.

A decision forecast evaluates an outcome under each option in a decision. For example, an agent can compare several grant amounts and estimate the resulting number of parliamentarians. The request can include context about the decision-maker, relevant facts, timing, money, and prior actions that are not available on the web.

## How it works

The hosted MCP endpoint is `https://mcp.futuresearch.ai/mcp`. An assistant sends forecast or research requests through that endpoint and receives structured output. Forecast operations work with tabular inputs, where each row contains a question and, when needed, fields such as alternatives, categories, thresholds, or conditions.

Supported forecast forms include binary questions, numeric estimates, dates, multiple-choice outcomes, threshold probabilities, and forecasts conditioned on a premise. Numeric and date forecasts return percentile estimates. Categorical forecasts return probabilities across the supplied options, while conditional forecasts provide results both given and not given the condition. Forecast results include a rationale column explaining the answer.

## Setup and configuration

Connect an MCP-compatible client to `https://mcp.futuresearch.ai/mcp`. The README specifically documents connections for Claude.ai, Claude Code, Gemini CLI, Codex CLI, and Cursor. The Python SDK is installed separately with `pip install futuresearch` and requires Python 3.12 or newer. Its API key is obtained from the FutureSearch application, and the SDK documentation notes that the older `everyrow` package is deprecated in favor of `futuresearch`.

Consult the client-specific connection instructions and FutureSearch documentation before deploying the connection.

## Tools and capabilities

- Forecast a binary yes/no question with a probability.
- Estimate numeric quantities using percentile outputs and units.
- Estimate dates using percentile dates in `YYYY-MM-DD` format.
- Compare categorical outcomes with jointly calculated probabilities.
- Evaluate threshold conditions or forecasts under a stated condition.
- Forecast decisions across alternatives, including decisions made by another party.
- Run `agent_map()` web research across every row of a dataset.
- Use `multi_agent()` research to build forecast inputs or answer non-forecast questions.

Forecasts can use low or high effort where the operation exposes that choice. Decision forecasts run at high effort, and categorical, thresholded, and conditional forecasts require high effort. Costs vary by operation and effort level; the README directs users to the current pricing information.

## Limitations and notes

Forecast quality and interpretation depend on clearly stated questions, options, context, and resolution criteria. Numeric and date requests require an output field, and numeric requests also require units. Categorical requests need a field containing the possible outcomes; the README recommends making that set exhaustive and adding an “Other” option when appropriate.

The material identifies the MCP service as hosted, but does not provide a local MCP server installation procedure. The Python SDK and MCP connection are related access paths, but the documented SDK commands should not be assumed to configure the hosted endpoint automatically.

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