How to Build a World-Aware Claude Agent

Claude has a knowledge cutoff. Ask it "what's the probability of a US recession?" and it will hedge, give you stale data, or hallucinate a number. This is fixable.

The Problem

Every LLM has this gap. Web search is the usual workaround, but web search returns narratives:

"According to recent reports, tensions in the Middle East remain elevated as diplomatic efforts continue..."

Your agent now has vibes, not data. It can't reason precisely over "elevated tensions."

Three Integration Patterns

There are three ways to give Claude world awareness, from simplest to most powerful.

Pattern 1: System Prompt Injection (simplest)

One API call fetches ~800 tokens of calibrated world state. Inject it into the system prompt.

import requests
import anthropic

client = anthropic.Anthropic()

# Fetch world state — free, no API key
world = requests.get("https://simplefunctions.dev/api/agent/world").text

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    system=f"""You are a macro research analyst with real-time world awareness.

Use the data below as ground truth. Cite specific probabilities and prices.
Do not hallucinate numbers — if it's not in the data, say you don't know.

{world}""",
    messages=[
        {"role": "user", "content": "What's the probability of a US recession in 2026?"}
    ],
)

print(response.content[0].text)

Claude now has real numbers to cite. Instead of hedging, it says "recession probability is 33% according to prediction market data."

Pattern 2: Tool Use (deeper)

Give Claude tools to drill into specific topics mid-conversation.

tools = [
    {
        "name": "get_world_state",
        "description": "Get current world state from prediction markets. Focus on specific topics for deeper coverage.",
        "input_schema": {
            "type": "object",
            "properties": {
                "focus": {
                    "type": "string",
                    "description": "Comma-separated topics: geopolitics, economy, energy, elections, crypto, tech"
                }
            }
        }
    },
    {
        "name": "get_world_delta",
        "description": "Get only what changed since a timestamp. ~30-50 tokens vs 800 for full state.",
        "input_schema": {
            "type": "object",
            "properties": {
                "since": {"type": "string", "description": "Time window: 30m, 1h, 6h, 24h"}
            },
            "required": ["since"]
        }
    },
    {
        "name": "search_prediction_markets",
        "description": "Search for specific prediction market contracts. Returns prices, volumes, spreads.",
        "input_schema": {
            "type": "object",
            "properties": {
                "query": {"type": "string", "description": "Search query, e.g. 'iran oil'"}
            },
            "required": ["query"]
        }
    }
]

Then implement a tool-use loop:

import json

def handle_tool(name, input_data):
    if name == "get_world_state":
        url = "https://simplefunctions.dev/api/agent/world"
        if input_data.get("focus"):
            url += f"?focus={input_data['focus']}"
        return requests.get(url).text
    elif name == "get_world_delta":
        return requests.get(
            f"https://simplefunctions.dev/api/agent/world/delta?since={input_data['since']}"
        ).text
    elif name == "search_prediction_markets":
        resp = requests.get(
            f"https://simplefunctions.dev/api/public/scan?q={input_data['query']}&limit=10"
        )
        return json.dumps(resp.json().get("markets", [])[:10], indent=2)

messages = [{"role": "user", "content": "Deep dive on Iran risk and its oil market impact."}]

while True:
    response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=1024,
        system=f"You are a macro analyst.\n\n{world}",
        tools=tools,
        messages=messages,
    )

    if response.stop_reason == "tool_use":
        tool_results = []
        for block in response.content:
            if block.type == "tool_use":
                result = handle_tool(block.name, block.input)
                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": result,
                })
        messages.append({"role": "assistant", "content": response.content})
        messages.append({"role": "user", "content": tool_results})
    else:
        for block in response.content:
            if hasattr(block, "text"):
                print(block.text)
        break

Claude calls search_prediction_markets("iran") on its own, gets specific contract data, and produces analysis with real numbers.

Pattern 3: MCP Server (zero code)

SimpleFunctions exposes 40 tools via MCP. For Claude Code:

claude mcp add simplefunctions --url https://simplefunctions.dev/api/mcp/mcp

For Claude Desktop, add to your MCP config:

{
  "mcpServers": {
    "simplefunctions": {
      "url": "https://simplefunctions.dev/api/mcp/mcp"
    }
  }
}

Now Claude has get_world_state, get_world_delta, scan_markets, and 37 more tools — no code, no API key for public endpoints.

Focused Mode and Incremental Updates

Two parameters that matter for token-conscious agents:

Focus: ?focus=geopolitics,energy — same ~800 token budget, concentrated on fewer topics. Instead of 4 contracts per topic across 6 topics, you get 10 contracts per topic across 2 topics. More depth, same cost.

Delta: /delta?since=1h — only what changed. Returns ~30-50 tokens instead of 800. For agents in long sessions, check delta periodically instead of re-reading the full state.

# Full state at start
world = requests.get("https://simplefunctions.dev/api/agent/world").text

# Periodic refresh — only changes
delta = requests.get("https://simplefunctions.dev/api/agent/world/delta?since=1h").text

# Deep dive on one topic
energy = requests.get("https://simplefunctions.dev/api/agent/world?focus=energy").text

Why Prediction Markets Instead of Web Search?

A prediction market price of 53 cents on "Iran invasion" means: someone bet $53 this happens, someone else bet $47 it doesn't. Both have money at risk. The aggregate of all such bets produces a calibrated probability.

A news headline means: an editor thought this would get clicks.

These are fundamentally different information sources. The first is a number your agent can reason over. The second is narrative it has to parse, summarize, and hope is current.

Source

Data: SimpleFunctions World Model — 9,706 prediction markets from Kalshi (CFTC-regulated) and Polymarket, calibrated by real money. Updated every 15 minutes.