> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wavestreamer.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Setup

> Connect Rayify to Claude, Cursor, ChatGPT, or any MCP-compatible AI client.

The Rayify MCP server (`@rayify-ai/mcp`) is the slim-8 integration surface designed for LLM-driven workflows. It exposes 4 pull tools + 4 push tools + 3 guided prompts.

## Install

The MCP server runs via `npx` - no install step required:

```bash theme={null}
npx -y @rayify-ai/mcp
```

## Configure your AI client

### Claude Desktop

Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%/Claude/claude_desktop_config.json` (Windows):

```json theme={null}
{
  "mcpServers": {
    "rayify": {
      "command": "npx",
      "args": ["-y", "@rayify-ai/mcp"],
      "env": {
        "RAYIFY_API_KEY": "sk_your_key"
      }
    }
  }
}
```

Restart Claude Desktop. The Rayify tools appear in the tool list with a hammer icon.

### Cursor

Open Cursor Settings → Features → Model Context Protocol → Add MCP Server. Configure:

* **Name:** `rayify`
* **Command:** `npx`
* **Args:** `-y @rayify-ai/mcp`
* **Env:** `RAYIFY_API_KEY=sk_your_key`

### ChatGPT (via custom GPT)

ChatGPT does not yet have native MCP support. Use the [TypeScript SDK](/sdk-setup#typescript) instead, exposed as a Custom Action.

### Generic MCP client

Any client that speaks MCP can connect:

```bash theme={null}
RAYIFY_API_KEY=sk_your_key npx -y @rayify-ai/mcp
```

The server speaks MCP over stdio.

## The 8 tools

### Pull (4)

| Tool                 | Purpose                                                                               |
| -------------------- | ------------------------------------------------------------------------------------- |
| `list_my_projects`   | List the workspace's projects, optionally filtered by status / kind                   |
| `get_project`        | Pull a project's full state (questions, reasonings, DecisionPacket) into your context |
| `get_project_status` | Light status check - pings the project without dragging full state                    |
| `search_findings`    | Semantic search across the workspace's project corpus                                 |

### Push (4)

| Tool                      | Purpose                                               |
| ------------------------- | ----------------------------------------------------- |
| `create_question`         | Add a question to an in-flight project                |
| `create_research_project` | Start a research project from a one-shot brief        |
| `attach_source`           | Attach a URL or document to a project's evidence pack |
| `create_survey`           | Run a list of questions as a multi-agent survey       |

## The 3 guided prompts

Slash-prompts that an MCP client can surface:

| Prompt         | What it does                                                                              |
| -------------- | ----------------------------------------------------------------------------------------- |
| `/pull-recent` | Lists the workspace's 10 most recent projects with their status                           |
| `/cite-rayify` | Wraps your current response with a Rayify citation block (links + DecisionPacket excerpt) |
| `/quick-ask`   | Starts a one-shot research project from your current chat context                         |

## What's not in the MCP surface

The MCP server is intentionally slim - heavy authoring lives on the web app (`rayify.ai`):

* Matrix / likert / star\_rating question authoring (`create_survey` accepts string arrays only)
* Project editing, deleting, sharing
* Agent template editing
* Org / billing / settings

If you need any of these, use the [SDK](/sdk-setup) or the web app.

## Authentication

The MCP server reads `RAYIFY_API_KEY` from the environment. Pass it via the `env` block of your MCP client config, never hardcoded.

For workspaces with multiple users, provision a per-user API key - the MCP server's actions are recorded under the key's owner in the audit log.

## What's next

* [SDK Setup](/sdk-setup) - for non-AI workflows
* [API Reference](/api-reference) - the underlying REST API