Get This Tool
Makoto
Pricing
- Model
- Free
Summary
Claude Code tells you the tests passed. You ship. Later you find out it never ran them. Makoto exists for that exact gap — cross-referencing every claim an agent makes against its own logged event record before anything moves forward.
Makoto hooks into Claude Code's event stream and audits each assertion — test results, citation matches, commit records, certificate claims — against a ledger of what the agent actually did, not what it reported. The vendor states the design targets zero false positives, meaning Makoto blocks on confirmed fakes rather than flagging on suspicion. That precision matters in CI gates where a noisy checker gets disabled within a week. The tool is reactive, not autonomous: it sits between agent action and downstream consequence, checking receipts. Teams without Claude Code in their stack have nothing to hook into — this is not a general-purpose verification layer.
Bottom line: Makoto is the right call for any Claude Code pipeline where a fabricated test result or citation would reach production unchallenged — it is the wrong call if your agent stack runs on anything other than Claude Code, because the hooks simply do not exist.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Blocks fabricated test and verification claims at the event level rather than logging them after the fact, so a false 'tests pass' report cannot reach a deployment gate unchallenged.
- Zero-false-positive design means the block signal stays meaningful — teams do not disable it after the first week of noise, which is what happens to checkers that flag on suspicion rather than confirmed mismatch.
- Apache-2.0 licensed and self-hosted, so the verification record never leaves your infrastructure — audit trails for compliance workflows stay under your control without a third-party dependency.
- Citation and commit record matching is built into dedicated modules, so agent-generated artifacts that cite sources or reference commits get cross-checked against actual logged activity, not just pattern-matched against formatting.
- Reactive hook architecture means Makoto adds a check layer without replacing the agent's planning or execution flow — you keep the Claude Code workflow intact and add the integrity gate on top.
Cons
Sign in to edit- The integration is Claude Code-specific with no API and no documented hook layer for other agent frameworks — teams running GPT-4 function-calling pipelines, LangGraph, or any non-Claude stack hit a dead end at installation and move to a custom audit logging layer instead.
- Self-hosted operation means your team owns the ledger storage, schema migrations, and the dispatch infrastructure — the `db.py` and `schema.py` files indicate local persistence you must manage, which becomes a maintenance burden when the project updates its schema and your production ledger does not.
- With zero open issues and zero pull requests on a 37-star repo, community-sourced fixes for edge cases in citation or commit matching are not available — teams that hit a verification gap in production write the patch themselves or file it and wait.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- Python, Claude Code
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-07-09T04:23:48.397Z
Best For
Who it's for
- Claude Code users needing claim verification
- Agentic workflows requiring zero-tolerance integrity
- Teams reviewing agent-generated artifacts
What it does well
- Verify test runs claimed by Claude Code
- Enforce citation and commit record matching
- Block false certificate or verification claims
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is Makoto free?
- Yes — Makoto is fully free to use. There is no paid tier.
- Is Makoto open source?
- Yes. Makoto is open source.
- Can I self-host Makoto?
- Yes. Makoto supports self-hosting on your own infrastructure.
- What platforms does Makoto support?
- Makoto is available on: Python, Claude Code.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
When an agent claims it completed a task, the claim and the deed are two separate things. Makoto sits as a hook layer inside Claude Code’s event architecture, capturing what the agent actually logged — commands run, commits made, citations sourced — and comparing that record against what the agent reported. If the records do not match, the action is blocked, not warned about. The source tree shows dedicated modules for checks, prechecks, stopchecks, citations, commitments, receipts, and a ledger, indicating a structured audit trail rather than ad-hoc string matching.
The distinguishing design choice, per the repo description, is measured zero false positives. Most integrity checkers in agent pipelines err toward sensitivity and produce noise that teams learn to ignore. Makoto’s stated posture is to block only on confirmed discrepancies, which means the block carries weight when it fires. The `ackblock.py` and `retraction.py` modules in the source suggest a workflow where blocked claims can be acknowledged or retracted through a defined path, rather than silently dropped.
Makoto fits tightly scoped Claude Code pipelines where fabricated outputs carry real cost — automated deployments, compliance artifacts, audit trails. It does not fit as a general agent monitoring layer: the integration is Claude Code-specific, there is no API surface, and teams running other agent frameworks will find no supported hook points. The project is Apache-2.0 licensed, self-hostable, and downloaded directly from the GitHub repository, so there is no dependency on an external service for the verification logic itself.
The `.claude-plugin` directory in the source tree confirms the integration mechanism is Claude Code’s plugin architecture. The `hooks` directory and `_dispatch_shim.sh` suggest event-driven dispatch rather than polling, which keeps verification latency out of the agent’s hot path. Teams adopting this should expect to instrument their Claude Code setup and maintain the ledger storage — the `db.py` and `schema.py` files indicate local persistence is part of the design.