rei-checker
rei-checker-mcp
形式验证检查器 MCP v0.1.0a1 — 接收一行,返回真伪。不多也不少。
三值判定(VALID / INVALID / UNDECIDED)。判定路径中不涉及 LLM。每个 UNDECIDED 都带有原因代码,并记录在仅追加的反驳账本中。
许可证:AGPL-3.0-or-later 规范:CHECKER_SPEC_v0.md 设计不变量:CLAUDE.md
5 分钟运行
有 Python 3.9+ 就足够了。无外部依赖(仅标准库)。
git clone https://github.com/fc0web/rei-checker-mcp.git
cd rei-checker-mcp
python -m rei_checker verify "1 + 1 = 2"预期输出:
{
"verdict": "VALID",
"elapsed_ms": 0,
"checker_version": "rei-checker-mcp/0.1.0a1+spike-2026-08-22"
}对于无法判定的输入,返回「无法判定」(spec §1.2):
python -m rei_checker verify "some unknown thing"{
"verdict": "UNDECIDED",
"elapsed_ms": 0,
"checker_version": "rei-checker-mcp/0.1.0a1+spike-2026-08-22",
"reason_code": "OUT_OF_SCOPE",
"detail": "MockBackend has no rule for this expression"
}退出码:0 = 决定性(VALID/INVALID),2 = UNDECIDED。可在 shell 脚本中直接根据「是否判定成功」进行分支。
Related MCP server: Chiasmus
Ledger 累积与 stats
所有 verify 调用都会在 ledger.jsonl 中追加一行(spec §4)。
python -m rei_checker verify "1 + 1 = 2"
python -m rei_checker verify "1 + 1 = 3"
python -m rei_checker verify "<axiom-test>"
python -m rei_checker stats{
"total": 3,
"valid": 1,
"invalid": 1,
"undecided": 1,
"decision_rate": 0.6666666666666666,
"reason_breakdown": {
"MISSING_AXIOM": 1
}
}decision_rate 是唯一指标(spec §3)。初始值为 0.1 也没关系 — 达到可测量的状态 才是成功条件。
Ledger 的位置可通过 $REI_CHECKER_LEDGER 环境变量覆盖。默认 = 当前目录下的 ledger.jsonl。
作为 MCP server 使用
注册到 Claude Desktop:
{
"mcpServers": {
"rei-checker": {
"command": "python",
"args": ["-m", "rei_checker", "mcp"],
"cwd": "C:/path/to/rei-checker-mcp",
"env": {
"REI_CHECKER_LEDGER": "C:/path/to/ledger.jsonl"
}
}
}
}MCP tool 只有 2 个(spec §2,刻意保持最小):
verify(expression, context?, timeout_ms?)→{ verdict, reason_code?, detail?, elapsed_ms, checker_version }stats()→{ total, valid, invalid, undecided, decision_rate, reason_breakdown }
哪些「尚未实现」(spec §2 明确)
以下为 v0 的非目标。如果试图实现,请 停下来确认:
UI / Web 前端
用户注册・认证・计费
游戏化、进度管理、学习记录
自然语言对话・解说生成
多后端支持(仅 Lean 4,v0 spike 使用 Mock backend 运行)
对 Claude 特有功能的依赖
v0 的状态(诚实范围,2026-08-22 spike)
✅ Schema(3 值 + 6 种 reason code)完全实现
✅ Mock backend(测试用真值表 + 所有 reason code 触发)
✅ Ledger(仅追加 JSONL、UTF-8、跳过格式错误的行)
✅ stats() 聚合(decision_rate + reason_breakdown)
✅ MCP stdio server(initialize + tools/list + tools/call)
✅ CLI(verify / stats / mcp / version 子命令)
⚠ Lean 4 backend 是 stub(v0.2 候选,计划在 lean_backend/ 目录实现)
⚠ Timeout 执行是软性的(监控 elapsed,硬性进程终止在 v0.2)
「先被使用」优先(spec §1.3、§6.6)。Lean 4 harness 完成后,替换 backend 即可实际判定运行。API surface 不变。
Phase 2(v0 完成前不开始)
spec §9-13 定义的三层结构:
第 1 层 checker(verify / stats)← v0,即本仓库
第 2 层 education(locate_first_error / boundary_report / escalate)
第 3 层 harness(calibration / regression / transfer)
实现顺序:第 1 层 → 第 3 层② calibration harness → 第 2 层。详见 spec §9-13。
测试
python -m unittest tests.test_all -v按照 Spec §7,优先测试应返回 UNDECIDED 的情况(所有 reason_code 单独测试 + VALID/INVALID 正常路径)。
与 Rei stack 的关系(避免混淆)
本仓库刻意保持独立。与相邻工具的区别:
rei-verify(PyPI 0.1.0a1)= 反证机械 4 值判定,以 refutation-first 为主轴。本仓库为 3 值 verification-first,设计哲学不同。
grounded-check = LLM 输出的引用 grounding check,属于另一领域。
rei-preregister = 预测 SHA256 seal,事前注册工具。
discovery-worker = 反例 hunter,属于另一层。
由于集成违反 spec §5「多后端支持非目标」,目前不进行集成。
贡献 / 报告
请遵守 spec §8 的 4 项原则:
犹豫是否添加功能时,不添加
犹豫是否返回「大概正确」时,返回 UNDECIDED
犹豫是否将理论暴露到 API 时,不暴露
不着急,慢慢来
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseAqualityDmaintenanceMCP server that gives small LLMs verified symbolic-math & logic tools.62Apache 2.0
- AlicenseNot gradedqualityAmaintenanceMCP server that gives LLMs access to formal verification via Z3 and SWI-Prolog, plus tree-sitter-based source code analysis. Translates natural language problems into formal logic using a template-based pipeline, verifies results with mathematical certainty, and analyzes call graphs for reachability, dead code, and impact analysis.89207Apache 2.0
- FlicenseNot gradedqualityCmaintenanceMCP server that exposes the DALI2-Agent-Brain symbolic verification system as tools, allowing MCP clients to submit reasoning problems for formal Prolog-based verification.
- AlicenseNot gradedqualityBmaintenanceA verification infrastructure and MCP server that specializes in refutation (negation) rather than generation, providing tools for counterexample search, Lean verification, and audit chains with a 4-value verdict system.MIT
Related MCP Connectors
Free MCP tools: the only MCP linter, health checks, cost estimation, and trust evaluation.
Conformance checker for MCP servers. Free, no key, verdicts recomputable and re-measured daily.
A paid remote MCP for hosted MCP server, built to return verdicts, receipts, usage logs, and audit-r
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
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/fc0web/rei-checker-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server