Install AgentBets MCP Server — Prediction Market Docs for AI Assistants

What is this?

The Model Context Protocol (MCP) lets AI-powered development tools pull in external documentation as context. When you install the AgentBets MCP server, your AI assistant gets direct access to our 153+ guides on building prediction market agents — no copy-pasting, no searching.

Ask your AI assistant “how do I authenticate with the Polymarket CLOB API in Python?” and it reads the answer directly from our Polymarket API guide. Ask “what is the difference between Moltbook and SIWE for agent identity?” and it reads the relevant AgentBets comparison.

Your AI assistant gains permanent, up-to-date access to the most comprehensive documentation on the agent betting stack anywhere on the internet.

Option 1: Hosted MCP Endpoint (Zero Install)

Point any MCP client directly at our hosted endpoint — no local setup, no packages, no Python.

Endpoint: https://api.agentbets.ai/mcp

Transport: Streamable HTTP (JSON-RPC 2.0 over POST)

This endpoint exposes:

• 8 resources — llms.txt, llms-full.txt, agents.json, glossary.json, research.json, recent.json, taxonomy.json, vig-data.json
• 6 tools — search_guides (keyword search across all pages), get_page (fetch any page as markdown), lookup_term (glossary lookup), get_vig_rankings (live sportsbook vig rankings by sport), recommend_sportsbook (personalized top-3 sportsbook picks based on skill, bet size, and sport), ask_question (AI-powered natural language answers using live odds and vig data)

Configure your MCP client

For any MCP client that supports HTTP transport, point it at:

https://api.agentbets.ai/mcp

For agent frameworks and coding assistants that accept an MCP server URL, this is all you need.

Quick test with curl

# Server info
curl https://api.agentbets.ai/mcp

# List available tools
curl -X POST https://api.agentbets.ai/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# Search guides
curl -X POST https://api.agentbets.ai/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search_guides","arguments":{"query":"polymarket api"}}}'

# Ask the AI a question (powered by live odds data)
curl -X POST https://api.agentbets.ai/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"ask_question","arguments":{"question":"Who has the lowest NFL vig?"}}}'

Option 2: Local MCP Server (via mcpdoc)

Run the MCP server locally using the mcpdoc package. This fetches documentation from agentbets.ai on demand.

Prerequisites

• Python 3.10+ with uvx installed (pip install uv then uvx is available)
• One of: Claude Desktop, Claude Code, Cursor, or Windsurf

Verify your setup:

uvx --version
# Should output: uv 0.x.x or similar

If uvx is not found, install uv first:

pip install uv

Install

All tools use the same MCP server package (mcpdoc). Pick your tool below.

Claude Code (CLI)

One command — no config file needed:

claude mcp add agentbets -- uvx mcpdoc --urls "AgentBets:https://agentbets.ai/llms.txt"

Or add to your project’s .mcp.json for per-project config:

{
  "mcpServers": {
    "agentbets": {
      "command": "uvx",
      "args": ["mcpdoc", "--urls", "AgentBets:https://agentbets.ai/llms.txt"]
    }
  }
}

Claude Desktop

Add to your claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "agentbets": {
      "command": "uvx",
      "args": ["mcpdoc", "--urls", "AgentBets:https://agentbets.ai/llms.txt"]
    }
  }
}

Restart Claude Desktop after saving.

Cursor

Add to .cursor/mcp.json in your project root, or to ~/.cursor/mcp.json for global access:

{
  "mcpServers": {
    "agentbets": {
      "command": "uvx",
      "args": ["mcpdoc", "--urls", "AgentBets:https://agentbets.ai/llms.txt"]
    }
  }
}

Reload the Cursor window (Cmd+Shift+P → “Developer: Reload Window”) after saving.

Windsurf

Add to ~/.windsurf/mcp.json:

{
  "mcpServers": {
    "agentbets": {
      "command": "uvx",
      "args": ["mcpdoc", "--urls", "AgentBets:https://agentbets.ai/llms.txt"]
    }
  }
}

Verify it works

After installing and restarting your tool, ask:

“Using the AgentBets docs, what are the four layers of the agent betting stack?”

Your assistant should respond with details about Identity, Wallet, Trading, and Intelligence layers — pulled directly from our documentation.

For a more specific test:

“How do I place a market order on Polymarket using py_clob_client?”

The assistant should return working Python code from our py-clob-client guide.

What gets served

The MCP server fetches llms.txt, which indexes all 153+ pages. Your AI assistant fetches individual pages on demand when you ask a relevant question. Content available:

The llms-full.txt file contains complete content for every page — useful for tools that prefer a single large context load.

Why install this?

Every time you ask your AI assistant a question about prediction market agents, it has our documentation in its context window. No Googling, no tab-switching — your AI reads our guides the same way it reads your codebase.

Example questions you can ask after installing

Architecture questions:

• “What are the four layers of the agent betting stack?”
• “How does agent identity work on prediction markets?”
• “What is the difference between Coinbase Agentic Wallets and regular wallets?”

API and code questions:

• “Show me how to authenticate with the Kalshi API in Python”
• “How do I get the order book for a Polymarket market using py_clob_client?”
• “What are the rate limits on the Polymarket CLOB API?”

Strategy questions:

• “How does cross-market arbitrage work between Polymarket and Kalshi?”
• “What is closing line value and how do I track it programmatically?”
• “How do I implement Kelly criterion position sizing in a betting agent?”

Tool and ecosystem questions:

• “What is Moltbook and how does it provide agent identity?”
• “What prediction market tools support automated order placement?”
• “What is the best wallet for a Polymarket trading agent?”

Sportsbook and vig questions:

• “Which sportsbook has the lowest vig on NFL games?”
• “What is the overall vig ranking across all US sportsbooks?”
• “How does BetOnline compare to Bovada on NBA vig?”

Troubleshooting

“MCP server failed to start”

Verify uvx is installed and accessible:

which uvx  # macOS/Linux
where uvx  # Windows

If not found, install uv:

pip install uv

“Tool not available” in Claude Desktop

Check that your JSON config is valid (no trailing commas, proper quotation marks). Test validity at jsonlint.com. Ensure you restarted Claude Desktop after editing the config file.

“No results from AgentBets”

The MCP server fetches https://agentbets.ai/llms.txt over the network. Ensure your machine has internet access. Try accessing https://agentbets.ai/llms.txt in a browser to confirm it loads.

Cursor: MCP config not being read

Cursor reads .cursor/mcp.json from the workspace root. Make sure you have a workspace folder open (not just a single file), and that the file is at the root of that workspace, not in a subdirectory.

Data access

The AgentBets MCP server exposes read-only access to all site data:

Text endpoints:

• /llms.txt — Structured index of all pages, organized by section with titles and URLs. Used as the MCP entry point.
• /llms-full.txt — Full text of all 153+ pages concatenated, with per-page metadata. Use this for tools that load full documentation context upfront.
• Per-page markdown — Any page can be fetched as clean .md by appending index.md to its URL (e.g. /guides/polymarket-api/index.md).

JSON APIs:

• /api/agents.json — Structured agent/tool directory with strategy, platform, and pricing data
• /api/glossary.json — 131 prediction market and agent betting terms
• /api/research.json — Ecosystem statistics and research data
• /recent.json — Recently added and updated pages with dates and summaries
• /taxonomy.json — Knowledge graph mapping pages to platforms, tools, tags, and layers
• /api/openapi.yaml — OpenAPI 3.1 spec describing all JSON endpoints
• /vig-data.json — Sportsbook vig/overround rankings — 11 US books across 8 sports with letter grades (A+ through F), trend indicators, and events sampled

All endpoints are public, free, and require no authentication.