Categories › AI & Agents › Multi-agent, orchestration & workflows
Independent signed math checks for AI agents: PASS or FAIL plus a link anyone can open. Free beta.
Remote server: your requests and data go to checkbooth.dev. Claude Code: claude mcp add --transport http checkbooth https://checkbooth.dev/mcp · Claude app: Settings → Connectors → Add custom connector.
The tools this server listed when we connected to it, in its own words.
discover_capabilities — List the whole Checkbooth tool chest with honest status tags (BUILT, COMING, ON REQUEST). Filter by words, family or status. Free.verify_math_or_tolerance — Independent, signed recompute of your numbers: formula, worst-case tolerance stack, power budget, Nyquist sampling, or rated range. Returns run_formula — Run one of our built-in physics formulas (sound speed, echo range, drop shock, field magnitude, thermal, timing and more) on YOUR inputs. Firun_chain — Run up to 8 checks or formulas in one call and get ONE signed bundle receipt. Each step is either a verify_math_or_tolerance argument objectquote — Exact credit price of any capability before you spend anything. Free.verify_report_receipt — Check that a Checkbooth receipt is genuine and unchanged. Free, no key needed.request_capability — Vote for a COMING capability, or describe one we do not list. Free; helps decide what opens next.seal_bundle_receipt — Verify 2 to 20 of your Checkbooth receipts and seal them into ONE signed design-review bundle with a single /check link.verify_trig_expression — Recompute a formula that uses trig (sin, cos, tan, asin, acos, atan; degree forms sind, cosd, tand, asind, acosd, atand) and compare it to aaudit_text_rounding — Paste text (up to 8,000 characters). We find every inline "arithmetic = number" claim and recompute it at the stated digits. Signed receipt.verify_claims_against_receipts — Paste a README, PR description, spec or memo plus the Checkbooth results it relies on. We check every number in the text against those signeissue_sandbox_key — Get a free capped beta key (cb_test_...). Shown once; we keep only its hash.Not tested by KitScout yet. When our harness runs it, pros, cons and the tasks we tried appear here, with the date and evidence.