thoughtproof-mcp
OfficialThe ThoughtProof MCP server verifies AI-generated reasoning and claims using adversarial multi-model consensus (Grok, Gemini, DeepSeek, Sonnet), helping you decide whether to trust and act on AI outputs.
Verify claims and reasoning: Submit any decision or reasoning claim via
verify_claim/verify_reasoningto receive a verdict (ALLOW, HOLD, UNCERTAIN, or DISSENT), a confidence score, and up to 3 key objections explaining why a claim may be challengedDomain-specific verification: Tailor analysis to
financial,medical,legal,code, orgeneralcontexts for more accurate assessmentsRisk-adjusted analysis: Set a stake level (
low,medium,high,critical) to adjust confidence thresholds based on decision consequenceAdjust verification depth: Choose between fast (2 models), standard (4 models), or deep (5+ models) to balance speed and cost ($0.008–$0.08 per verification)
Check agent trust scores: Use
check_agent_scoreto look up composite trust scores for specific agents, optionally filtered by domainGuard against hallucinations: Use verdicts and objections to validate AI outputs before acting on them
thoughtproof-mcp
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.
Tools
verify_claim
Verify any claim or AI-generated reasoning before acting on it.
Parameter | Type | Default | Description |
| string | (required) | The text to verify |
| low / medium / high / critical | medium | Risk level |
| financial / medical / legal / code / general | general | Domain context |
| fast / standard / deep | standard | Verification depth |
check_agent_score
Look up an agent's composite trust score.
Parameter | Type | Description |
| string | Agent ID to look up |
| string | Optional domain filter |
Setup
Claude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"thoughtproof": {
"command": "npx",
"args": ["-y", "thoughtproof-mcp"],
"env": {
"THOUGHTPROOF_API_KEY": "tp_op_your_key_here"
}
}
}
}Cursor
Add to Cursor's MCP settings:
{
"thoughtproof": {
"command": "npx",
"args": ["-y", "thoughtproof-mcp"],
"env": {
"THOUGHTPROOF_API_KEY": "tp_op_your_key_here"
}
}
}Windsurf / Cline
Same pattern — point to npx thoughtproof-mcp with the env var.
Example
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 ConsensusAPI Key
Get an operator API key at thoughtproof.ai. Without a key, verifications require x402 micropayment (USDC on Base).
Pricing
Speed | Models | Price |
fast | 2 | $0.008 |
standard | 4 | $0.02 |
deep | 5+ | $0.08 |
Development
npm install
npm run dev # Run with tsx
npm run inspect # Test with MCP Inspector
npm run build # Compile TypeScriptLicense
MIT — ThoughtProof
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/ThoughtProof/thoughtproof-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server