How much are you wasting on tokens?

Pick the sources your agent monitors. See exactly what raw polling costs vs. DiffDelta's cursor-based approach. Real data, real math.

Select sources
Monthly savings
0%
fewer tokens
$0.00
saved per month on LLM costs (GPT-4o pricing)
Sources selected 0
Without DiffDelta (tokens/mo) 0
With DiffDelta (tokens/mo) 0
Your cost without DiffDelta $0.00
Your cost with DiffDelta $0.00
Checks/day (all sources) 0

Your MCP config — paste into Cursor, Cline, or any MCP client

{
  "mcpServers": {
    "diffdelta": {
      "command": "npx",
      "args": ["-y", "@diffdelta/mcp-server@latest"]
    }
  }
}

Verify it yourself

Pick any source above and compare the raw upstream size to DiffDelta's head pointer:

# Head pointer (what your agent checks every cycle — ~570 bytes) curl -s https://diffdelta.io/diff/nist_nvd/head.json | wc -c # Full feed (what you'd parse without DiffDelta — ~213KB) curl -s https://diffdelta.io/diff/nist_nvd/latest.json | wc -c