@rayify-ai/mcp

MCP server for Rayify — connect your Rayify research projects to Claude Desktop, Cursor, ChatGPT, and any MCP-compatible client.

Pull findings into your AI workflow. Push new questions, sources, and context back into Rayify. Search your project corpus inline.

The web app (Ray + canvas at rayify.ai) remains the primary authoring UX. This MCP server is the integration surface — it makes the artifacts you produce in Rayify available wherever your AI lives.

Install

Claude Code

claude mcp add rayify -- npx -y @rayify-ai/mcp

Cursor / Claude Desktop / Windsurf

Add to your MCP config JSON:

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

Get your API key

Sign up at rayify.ai → Profile → Settings → API Keys.

What you can do

Pull from Rayify (4 tools)

| Tool | Use | |---|---| | list_my_projects | "What's in my Rayify workspace?" | | get_project | "Pull this diligence into my memo" | | get_project_status | Light polling without dragging full results into context | | search_findings | "Has Rayify ever concluded anything about X?" |

Push to Rayify (4 tools)

| Tool | Use | |---|---| | create_question | "Ask Rayify whether X — here's my context" | | create_research_project | "Run a research brief on these 12 URLs" | | attach_source | "Add this article to my in-flight project" | | create_survey | "Run this list of questions as a multi-agent survey" |

Guided prompts (3)

• /pull-recent — list your most recent projects + pull full details
• /cite-rayify — semantic-search your findings + format as inline citations
• /quick-ask — ask a question + poll for the answer

Example: pulling Rayify findings into a memo

In Cursor / Claude Desktop, after installing:

Me: "What did Rayify find about enterprise AI framework adoption?"

Claude: calls search_findings(query="enterprise AI framework adoption")

Claude: "Your Rayify projects have 3 relevant findings:

1. LangChain holds 38% but LangGraph adoption is accelerating
2. Enterprise teams cite stability concerns as #1 blocker ..."

Auth

API key is read from RAYIFY_API_KEY env var at MCP startup. No in-tool registration — get the key from rayify.ai.

For HTTP transport (Smithery / streaming), the key is also accepted as ?RAYIFY_API_KEY=... query string.

What's NOT in this surface (and why)

• Heavy authoring — custom question types (matrix, likert, star-rating) and rich canvas configuration stay in the web app. MCP JSON is the wrong tool for 50-row survey schemas.
• Project edits / deletes / sharing — web app.
• Leaderboard / follow / votes / personas / agent-registration — these surfaces were retired with the v2 product pivot.

Versioning

v2.0.0 introduces the slim 8-tool surface. v1.x's 71-tool surface targeted the retired AI-agent-leaderboard product and is no longer maintained.

Single source of truth: the VERSION file at the waveHub repo root.

Repo

Part of waveHub — Rayify's open-source SDKs, MCP server, and runner.

License: MIT.