Agent-native install

MCP Server

Drop-in Model Context Protocol server. Your AI agent reads Signal Bench data as first-class tools — no glue code, no SDK to learn.

What It Is

Signal Bench ships an official MCP (Model Context Protocol) server that any MCP-compatible AI agent can install. Claude Desktop, Cursor, Cline, Continue, and any custom agent built on the MCP standard can talk to Signal Bench through it. The agent gets every API endpoint as a discoverable, typed tool — no manual API wiring required.

Why this matters

Signal Bench is agent-native by design. The agent does the consumption and the interpretation; the user never types a Signal Bench query directly. Install the MCP server once, then your agent just knows when to fetch the daily digest, the latest regime, or the surprise index.

Install in Claude Desktop

Get a free Signal Bench API key from /settings/api-keys
Open your Claude Desktop config file — macOS: ~/Library/Application Support/Claude/claude_desktop_config.json — Windows: %APPDATA%\Claude\claude_desktop_config.json
Add the entry below under mcpServers
Restart Claude Desktop. Signal Bench tools appear in the tools menu.

{
  "mcpServers": {
    "signalbench": {
      "command": "npx",
      "args": ["-y", "@signalbench/mcp-server"],
      "env": {
        "SIGNALBENCH_API_KEY": "sb_yourkeyhere"
      }
    }
  }
}

Install in Cursor

Cursor → Settings → Features → Model Context Protocol → Add new MCP server. Paste:

{
  "command": "npx",
  "args": ["-y", "@signalbench/mcp-server"],
  "env": {
    "SIGNALBENCH_API_KEY": "sb_yourkeyhere"
  }
}

Install in Cline / Continue / Custom Hosts

Any host that supports stdio-transport MCP servers can run this binary. The invocation is the same: command is npx, args are -y @signalbench/mcp-server, and SIGNALBENCH_API_KEY must be in the env.

For environments without internet access during install, you can pre-install the package globally:

npm install -g @signalbench/mcp-server
SIGNALBENCH_API_KEY=sb_yourkeyhere signalbench-mcp

Available Tools

Tool	Returns	Tier
signalbench_list_signals	Full catalog of composite + momentum signals with latest values	Free
signalbench_get_signal	Single-signal detail (definition, history, drivers, validation)	Free
signalbench_get_digest	Today's biggest movers, regime shifts, warnings	Free
signalbench_get_alerts	Active threshold breaches, regime shifts, deteriorations	Standard
signalbench_get_regimes	Overall risk posture + per-category regime scores	Standard
signalbench_get_surprise_current	US Macro / Growth / Inflation trend-surprise indices	Standard
signalbench_get_surprise_history	Historical surprise timeseries	Standard
signalbench_get_relationships	Cross-market correlation pairs	Standard
signalbench_get_health	Data freshness, ingestion + validation status	Standard
signalbench_get_manifest	Full API manifest for further discovery	Free

Tier enforcement happens at the Signal Bench API; the MCP server passes your key through and surfaces any tier errors back to the agent as readable text.

Example Agent Prompts

Once installed, your agent can answer prompts like these without you ever touching the Signal Bench UI:

"What's the current US macro regime?"
"Did the equity regime flip today? If so, what drove it?"
"Summarize the macro surprise index for the past two weeks."
"Which composites deteriorated in this morning's digest?"
"Show me the credit stress composite's recent history and tell me if it's elevated."
"Are there any active high-severity alerts I should know about?"

The agent picks the right tool, the tool returns JSON, the agent explains it to you in its own voice. You get a daily macro read without ever opening a dashboard.

Configuration

Env var	Default	Purpose
SIGNALBENCH_API_KEY	(required)	Your Signal Bench API key
SIGNALBENCH_BASE_URL	https://www.signalbench.dev	Override for testing against a preview deployment

Source & Issues

The MCP server is open source under MIT. Source lives in the main signal-bench repo under packages/mcp-server/. File issues against the repo.

# clone + build locally if you want to hack on it
git clone https://github.com/rs25-labs/signal-bench
cd signal-bench
npm install
npm run mcp:build