Prediction Market X Sentiment

What Prediction Market X Sentiment does

Prediction Market X Sentiment MCP server exposes sentiment data for Polymarket and Kalshi markets based on live discussion on X, formerly Twitter. The service uses Grok to produce sentiment scores and supports both free daily data and paid live analysis. Its results are intended for agents that need to compare market prices with social attention or sentiment.

The free /scores route returns the latest daily X-sentiment score for each tracked market. Discovery endpoints, including /, /health, /sample, /catalog, and /pricing, are also free. Paid endpoints provide more current or focused analysis: /top identifies the three markets with the most X buzz, /shift?q= reports movement since the previous score, /sentiment?q= calculates a fresh score for a supplied question, and /brief?q= combines sentiment, change, a summary, and current Polymarket or Kalshi odds.

How it works

The hosted service is available at https://prediction-bot-iggf.onrender.com. Agents can connect to its /mcp endpoint as a remote, streamable-HTTP MCP server. A local MCP package is also available through uvx under the command prediction-bot-mcp.

Free tools do not require a wallet. Paid requests use HTTP 402 payment handling and settle in USDC on Base. The service publishes current prices at /pricing; the README lists default route pricing, but clients should query that endpoint rather than assume prices remain unchanged. The listed paid routes include separate pricing for lightweight requests such as buzz and shift data, and for fresh sentiment or brief responses.

The Prediction Market X Sentiment MCP server can be probed without payment through the health, catalog, and pricing endpoints. These routes help an agent or operator confirm availability, inspect the service catalog, and retrieve current pricing before requesting paid data.

Setup and configuration

For hosted use, add https://prediction-bot-iggf.onrender.com/mcp as a remote MCP server in a compatible client. No local installation is required for this option. For local use, configure an MCP server that runs uvx with the prediction-bot-mcp argument. The repository points to separate documentation for wallet configuration and per-call or per-session spending caps.

The deployment requires XAI_API_KEY for Grok access. Payment-enabled deployments use PAY_TO_ADDRESS as the USDC recipient on Base. Other documented settings control route prices, the public service URL, Grok spending, caller limits, and hourly request caps. Operators should review the repository's MCP documentation for wallet setup and cap behavior before enabling automatic payment.

Tools and capabilities

The Prediction Market X Sentiment MCP server supports these data functions through its HTTP API and MCP access:

• Retrieve the latest daily sentiment scores for tracked markets.
• Rank the top three markets by X buzz.
• Check sentiment movement since the previous score.
• Request a fresh Grok score from -100 to 100 for a market question.
• Get a combined score, shift, summary, and live Polymarket or Kalshi odds.
• Inspect service health, available routes, samples, and current pricing.

The service belongs in the Finance & Fintech category and is most relevant when social conversation is an input to prediction-market research. It does not replace the market data or payment configuration required by the service itself.

Limitations and notes

Live analysis is not uniformly free. Daily scores and discovery routes are free, while the live and ranked endpoints may return HTTP 402 until payment is completed. Payment requires USDC on Base and, for local automatic payment, a configured wallet and spending limits.

Sentiment is derived from X chatter through Grok, so the output represents social sentiment rather than a guaranteed forecast or a direct market consensus. Scores may change as new posts and market data are processed. Prices are configurable and may differ from the defaults documented in the README; /pricing is the authoritative source.

The repository includes a test command using TEST_MODE=1 pytest -q, but the provided material does not specify a published software license.