MCP server for ThoughtProof — verify AI reasoning with adversarial multi-model consensus.
4 LLMs (Grok, Gemini, DeepSeek, Sonnet) challenge each other on every claim. Returns ALLOW, BLOCK, or UNCERTAIN with confidence score and objections.
Verify any claim or AI-generated reasoning before acting on it.
| Parameter | Type | Default | Description |
|---|---|---|---|
claim |
string | (required) | The text to verify |
stakeLevel |
low / medium / high / critical | medium | Risk level |
domain |
financial / medical / legal / code / general | general | Domain context |
speed |
fast / standard / deep | standard | Verification depth |
Look up an agent's composite trust score.
| Parameter | Type | Description |
|---|---|---|
agentId |
string | Agent ID to look up |
domain |
string | Optional domain filter |
Add to your claude_desktop_config.json:
{
"mcpServers": {
"thoughtproof": {
"command": "npx",
"args": ["-y", "thoughtproof-mcp"],
"env": {
"THOUGHTPROOF_API_KEY": "tp_op_your_key_here"
}
}
}
}Add to Cursor's MCP settings:
{
"thoughtproof": {
"command": "npx",
"args": ["-y", "thoughtproof-mcp"],
"env": {
"THOUGHTPROOF_API_KEY": "tp_op_your_key_here"
}
}
}Same pattern — point to npx thoughtproof-mcp with the env var.
In Claude Desktop or Cursor, just ask:
"Verify the claim: GPT-5 achieves 95% accuracy on MMLU-Pro"
The tool returns:
⚠️ UNCERTAIN (42% confidence)
Claim: "GPT-5 achieves 95% accuracy on MMLU-Pro"
Objections:
- Insufficient public benchmark data to confirm
- Historical accuracy claims have been overstated
- MMLU-Pro methodology has known ceiling effects
⚡ 3.2s | Adversarial Multi-Model Consensus
Get an operator API key at thoughtproof.ai. Without a key, verifications require x402 micropayment (USDC on Base).
| Speed | Models | Price |
|---|---|---|
| fast | 2 | $0.008 |
| standard | 4 | $0.02 |
| deep | 5+ | $0.08 |
npm install
npm run dev # Run with tsx
npm run inspect # Test with MCP Inspector
npm run build # Compile TypeScriptMIT — ThoughtProof