Change Trace MCP
A local-first, model-neutral MCP server for collecting and normalizing change-scoped release evidence. It provides deterministic Git change summaries, evidence collection, and review bundles for agent review.
README
Change Trace MCP
Change Trace MCP is a local-first, model-neutral MCP server for collecting and normalizing change-scoped release evidence. The user's existing Agent makes semantic judgments; this package keeps evidence preparation deterministic and reviewable.
M1 Host and cloud-runner compatibility and the M2 deterministic evidence core are complete. M3 Agent review-contract work is in progress. The currently exposed MCP tools are:
get_server_inforeports process and runtime metadata;get_compatibility_fixturereturns a byte-stable fixture for Host smoke tests;get_change_scopereturns a deterministic, bounded Git change summary for an explicit repository root and two refs;collect_local_evidencereturns bounded, provenance-rich excerpts from configured repository document roots;get_review_bundlecombines change and document evidence into a bounded, indexed bundle with deterministic facts and missing-evidence records;validate_findingsvalidates Agent output against the shared schema and the bundle's evidence/source indexes;write_reportrenders validated findings as a deterministic Markdown and JSON report pair inside a repository-relative output directory.
Requirements
- Node.js 22 or newer.
Local development
npm install
npm run check
npm test
Run the stdio server from a local checkout:
npm run build
node dist/cli.js
The server reserves stdout for MCP JSON-RPC messages. Structured operational logs are emitted to stderr.
write_report usage
write_report accepts a validated ReviewBundle, a FindingValidationResult,
reviewer metadata, and a repository-relative output path. It renders a
deterministic Markdown report and a structured JSON report as a pair of files:
<reportName>.md— human-readable Markdown with safe code fences, escaped HTML, and bounded structure;<reportName>.json— machine-readable report conforming to the versionedReportJSON Schema.
The tool requires:
repositoryRoot— an absolute path to the repository working directory;outputDirectory— a path relative torepositoryRoot. Absolute paths,..traversal,.gitmetadata paths, and symlink escapes are rejected;reportName— a safe basename ([A-Za-z0-9][A-Za-z0-9._-]{0,127}) that determines the output filenames;bundle— a validatedReviewBundlefromget_review_bundle;validationResult— the matchingFindingValidationResultfromvalidate_findings. The validation result'sbundleIdmust equal the bundle's ID;reviewMeta— caller-supplied reviewer identity, requiredcreatedAttimestamp (ISO 8601), optional tool version, notes, and declared limitations; identical inputs includingcreatedAtproduce byte-identical reports;
Optional parameters:
overwrite— whentrue, existing report files are replaced. The default (false) refuses the write if either report file exists;maxReportSizeBytes— an output size bound. The default is 10 MiB; the absolute hard maximum is 100 MiB. The write fails instead of silently truncating findings.
The tool returns a structured result with reportId, absolute paths to the
written files, and their byte sizes. Error responses are bounded and do not
expose report content.
Example MCP call flow:
get_change_scope → collect_local_evidence → get_review_bundle →
validate_findings → write_report
Contribution workflow
This repository uses a coordinator/worker model with isolated worktrees,
tracked task assignments, and coordinator-owned Roadmap and release decisions.
See docs/CONTRIBUTING_WORKFLOW.md before
starting delegated implementation work.
Versioned schemas
The package exports strict Zod schemas and deterministic Draft 2020-12 JSON
Schema documents for EvidenceItem, ChangeScope, LocalEvidenceCollection,
ReviewBundle, Finding, FindingValidationResult, and Report:
import {
evidenceItemSchema,
exportCoreJsonSchemas,
} from "change-trace-mcp";
const evidence = evidenceItemSchema.parse(input);
const jsonSchemas = exportCoreJsonSchemas();
Host-specific setup and the current compatibility matrix live in
docs/smoke-tests/.
See docs/ROADMAP.md for milestone scope and
docs/PROJECT_DECISIONS.md for accepted product
and architecture decisions.
License
Apache-2.0. See LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。