Product
TPMN Checker
Verification for AI reasoning. 12 MCP tools, 6 domains, 3 providers.
Audits AI outputs for overclaims, extrapolation, and reasoning violations. Works in Claude.ai, ChatGPT, Claude Code, Cursor, and more.
Getting started
Connect via Claude.ai, ChatGPT, or install in your IDE.
Claude.ai Connector
- 1.Go to
claude.ai/customize/connectors - 2.Click “Add custom connector”
- 3.Name:
GEM2 TPMN CHECKER - 4.URL:
https://mcp-tpmn-checker.gemsquared.ai - 5.Complete OAuth login — enter LLM API keys on the consent page
- 6.Tools appear in your Claude.ai conversation
ChatGPT Connector
New- 1.Enable Developer Mode in ChatGPT Settings
- 2.Go to Apps & Connectors → Add App
- 3.Name:
GEM2 TPMN CHECKER - 4.URL:
https://mcp-tpmn-checker.gemsquared.ai/mcp - 5.Auth: OAuth → Complete login
- 6.Tools appear in ChatGPT
Claude Code / Cursor / IDE
npx (recommended)
npx @gem_squared/setupcurl
curl -sSL https://user-mgmt.gemsquared.ai/setup/install | bashAuto-detects AI tools, installs launcher, registers MCP server. Full setup guide →
Verify setup
Ask Claude or ChatGPT to run gem2_account_info to confirm your email, tier, configured providers, and default model.
What TPMN Checker detects
Instead of blindly accepting AI output, TPMN Checker audits the reasoning itself. It identifies problems that are difficult to catch because the reasoning sounds plausible.
TPMN Checker transforms AI usage from guess-and-trust into verify-then-use.
12 MCP tools
Grouped into three categories. Free tools cost no credits. Analysis tools use LLM cost + 1 credit per call.
Free — Info & Account (4 tools)
gem2_account_info
Show your email, tier, provider keys, and default model.
gem2_list_models
List available LLM models per provider from the live catalog.
gem2_drift_report
Show scoring calibration drift per LLM provider.
gem2_my_skills
List or search your stored prompts and contracts.
Analysis — LLM cost + 1 credit (4 tools)
gem2_truth_filter
Evaluate epistemic quality. Detect overclaims, extrapolation, and prompt-output alignment.
tpmn_p_check
Pre-flight content check with domain-specific verification. Strict, refine, or interpolate mode.
tpmn_p_check_compose
Rewrite content based on a Phase A diagnosis. Before/after truth scores.
tpmn_contract_writer
Generate a formal contract + grounded replacement prompt from a failed LLM interaction.
Skills — Consent-gated (4 tools)
gem2_store_prompt
Store a grounding prompt (5W context) as a reusable skill.
gem2_store_contract
Store a formal contract (JSON specification) as a reusable skill.
gem2_get_skill
Retrieve the full content of a stored skill by its ID.
gem2_skill_formatter
Reformat a skill for Claude Skill, GEM2 uniform, or raw-spec format.
6 supported domains
Domain is auto-detected from your content, or set manually with domain_override.
SDLC
Architecture, security, and performance claims
Medical
Source grounding, outcome qualification, risk disclosure
Research
Citations, methodology, statistical validity
Legal
Jurisdictional scope, statutory accuracy, precedent
Finance
Forward-looking disclosure, historical qualification
Marketing
Claim substantiation, comparative fairness
Common workflows
Diagnose → Compose
- 1.tpmn_p_check — get diagnosis + phase_a_id
- 2.tpmn_p_check_compose — rewrite with issues fixed
Or use compose=true for one-shot.
Interpolate → Research → Refine
- 1.tpmn_p_check (mode=interpolate) — get search suggestions
- 2.Research the suggested topics
- 3.tpmn_p_check (mode=refine) — validate grounded version
Best for claims that need external evidence.
Fix Failed Interaction
- 1.gem2_truth_filter — diagnose with prompt alignment
- 2.tpmn_contract_writer — get contract + replacement prompt
- 3.Use the replacement prompt going forward
Auto-stores as reusable skills when content logging is enabled.
Credits & tiers
Start free. Upgrade when you need more.
| Tier | Daily | Total Credits | RPM | Price |
|---|---|---|---|---|
| Free | 30 | 300 | 5 | $0 |
| Starter | ∞ | 2,000 | 15 | $9 |
| Builder | ∞ | 9,000 | 30 | $19 |
| Founder | ∞ | ∞ | 60 | $29 |
| Enterprise | ∞ | ∞ | 120 | Custom |
Pre-GA pricing. One-time payment, not subscription. Pricing and credits subject to change. Existing purchased credits remain valid. Get early access →
Real verification examples
Original AI output
"The fourth year of a bull market is historically positive."
TPMN Checker result
Issue: Unsupported historical claim
Action: Removed
Original AI output
"April is typically a high-stakes month."
TPMN Checker result
Issue: Overgeneralization
Action: Refined language
Demo video — coming soon
Two real verification cases showing before/after results