EngiMCP
An MCP server that gives LLM agents structured, safe, and traceable access to engineering project documentation stored in Markdown/Git repositories, enabling management of requirements, decisions, tests, tasks, and impact analysis.
README
id: README kind: overview status: draft version: 1.0.0
EngiMCP
EngiMCP is a local MCP server for managing engineering projects stored as Markdown, YAML frontmatter, and Git history. It gives an LLM agent structured, safe access to requirements, engineering decisions, calculations, tests, tasks, BOM items, document relationships, impact analysis, context packs, and ordinary project files.
The source of truth stays in the project folder. The SQLite index under .engimcp/index.sqlite is derived data and can be rebuilt.
Why It Exists
Long-running engineering work cannot live only in chat. Requirements drift, decisions get forgotten, tests stop matching design assumptions, and documents become inconsistent.
Codex is strong at inspecting and changing code repositories, but engineering documentation needs a domain layer on top of files. EngiMCP exposes IDs, document kinds, frontmatter, links, graph relationships, validation, impact analysis, and focused writes so the agent can answer questions such as:
What breaks if we change the battery?
Prepare context for changing the project motors.
Create an EDR for this design choice.
Find requirements that are not verified by tests.
Requirements
- Node.js 20 or newer.
- npm.
- Git, for Git status/snapshot/commit workflows.
- An MCP-compatible client such as Codex, Claude Code, or another local MCP client.
User Guide
EngiMCP is meant to be used through an MCP client. In normal use, you do not call TypeScript code directly. You start the server, connect it to your agent, and then ask the agent engineering-project questions in plain language.
1. Prepare an Engineering Project Folder
Choose or create a local folder for the project:
mkdir -p ~/Work/Projects/RC-Car
cd ~/Work/Projects/RC-Car
git init
EngiMCP will keep the source of truth in this folder as Markdown files plus project.yaml.
2. Install and Build EngiMCP
From the EngiMCP repository:
npm install
npm run build
3. Connect It to an MCP Client
Use the built server command in your MCP client configuration:
node /absolute/path/to/EngiMCP/dist/index.js --root /absolute/path/to/your/project
For a read-only review session:
node /absolute/path/to/EngiMCP/dist/index.js --root /absolute/path/to/your/project --read-only
After package installation, the binary form is:
engimcp --root /absolute/path/to/project
engimcp --root /absolute/path/to/project --read-only
When --root is set, tool calls must use that same project root. When --read-only is set, write-like tools are rejected.
4. Initialize the Project
Ask your agent:
Initialize this folder as an EngiMCP project.
The agent should call:
engi_project_init(root="/absolute/path/to/project")
This creates the basic project files:
project.yaml
docs/
templates/
.engimcp/
5. Start Working
Useful first prompts:
Show project status.
Create a requirement: the car must run for 30 minutes.
Create an engineering decision for using four geared motors.
What breaks if we change the battery?
Prepare context for changing the project motors.
Find requirements that are not verified by tests.
Validate the project and show what needs fixing.
Show the Git diff summary.
The agent will use EngiMCP tools to read the right documents, build context, analyze links, and apply focused edits.
Common User Workflows
Create Requirements
Ask:
Add a requirement that the radio-controlled car must stop safely when radio control is lost.
EngiMCP creates a managed Markdown requirement with an ID such as FR-001, NFR-001, SEC-001, or AC-001.
Record Engineering Decisions
Ask:
Create an EDR explaining why we selected a 2S LiPo battery.
EngiMCP creates an Engineering Decision Record such as EDR-0001 with context, options, decision, consequences, and links.
Analyze Change Impact
Ask:
What documents, tests, BOM items, and decisions are affected if we change the battery?
EngiMCP follows frontmatter links and inline links to return direct and transitive impact.
Prepare Focused Context
Ask:
Prepare context for changing the drive motors.
EngiMCP builds a compact context pack instead of forcing the agent to read the whole project.
Edit Documents Safely
Ask:
Update the Battery Estimate section with the new runtime calculation.
The agent should use section-level document tools, so the edit stays focused and reviewable in Git.
Work with Ordinary Files
Ask:
List files under docs.
Read docs/battery-notes.md.
Create a folder for test reports.
Move this draft file into docs/archive.
Delete this temporary note.
EngiMCP uses the safe engi_fs_* layer. Deletes go to .engimcp/trash/... by default.
Project Format
A valid project has:
project.yaml
docs/
templates/
.engimcp/
Managed Markdown documents use YAML frontmatter with at least:
---
id: DOC-EXAMPLE
kind: design_doc
status: draft
version: 0.1.0
---
You can edit Markdown files manually in VS Code, Obsidian, or another editor. Run validation after manual edits:
Validate the project.
Quick Start
- Start the MCP server from this repository or install it as a binary.
- Connect an MCP client over stdio.
- Initialize or open a project root.
- Call
engi_project_statusto verify the project. - Use
engi_context_packbefore asking the agent to change a subsystem. - Use
engi_impact_analyzebefore changing requirements, decisions, battery, motors, BOM, tests, or calculations. - Use
engi_validate_projectafter changes. - Review the Git diff with
engi_git_status.
Example workflow:
engi_project_status(root)
engi_context_pack(root, task="we are changing the project motors")
engi_impact_analyze(root, changed_ids=["DOC-RC-CAR-BATTERY"])
engi_doc_patch_section(root, id="DOC-RC-CAR-BATTERY", heading_path=["Initial Estimate"], ...)
engi_validate_project(root)
engi_git_status(root)
Core Tools
Project:
engi_project_initengi_project_statusengi_project_mapengi_rebuild_index
Managed engineering documents:
engi_doc_readengi_doc_createengi_frontmatter_patchengi_doc_patch_sectionengi_doc_add_relationship
Engineering entities:
engi_requirement_createengi_decision_createengi_task_createengi_test_report_createengi_bom_item_create
Analysis and retrieval:
engi_searchengi_graph_queryengi_impact_analyzeengi_context_packengi_validate_project
Git:
engi_git_statusengi_project_snapshotengi_git_commit
Safe filesystem layer:
engi_fs_treeengi_fs_listengi_fs_readengi_fs_writeengi_fs_mkdirengi_fs_moveengi_fs_copyengi_fs_deleteengi_fs_existsengi_fs_statengi_fs_glob
Use engi_doc_* for managed engineering documents where IDs, frontmatter, links, graph, and validation matter. Use engi_fs_* for ordinary files and folders inside the project root.
Safety Model
EngiMCP is not raw filesystem access.
- All operations are restricted to
project_root. - Runtime
--rootpins the allowed project root. ../traversal and absolute-path escape are rejected.- Symlink escape outside the root is rejected.
- Deny patterns block
.git,.env, SSH keys, secrets, private files, cache paths, and dependency folders. - Project-configured deny patterns extend the defaults.
- Write-like tools respect read-only mode.
- Deletes use trash by default:
.engimcp/trash/YYYY-MM-DD/<original-path>. - Writes are atomic where practical.
- Write/move/copy/delete/mkdir operations write audit log entries.
- Overwrite-like operations refuse to modify files containing Git conflict markers.
Validation and Indexing
engi_validate_project checks IDs, required frontmatter, duplicate IDs, broken links, dependency cycles, and requirement verification gaps.
engi_rebuild_index regenerates .engimcp/index.sqlite from Markdown and frontmatter. Filesystem write-like operations rebuild the index after successful changes, and engi_project_status restores the index if it is missing.
Development Commands
npm run build
npm run lint
npm run format
npm run format:check
npm test
npm run check
npm run smoke
npm run check runs build, lint, format check, and the test suite. npm run smoke builds the project and runs a project-status smoke check against the radio-controlled car example.
Important Documentation
docs/02_requirements/technical_requirements.md- functional and non-functional requirements.docs/02_requirements/filesystem_overlay_requirements.md- safe filesystem layer requirements.docs/02_requirements/acceptance_criteria.md- acceptance criteria.docs/03_architecture/system_architecture.md- system architecture.docs/04_data_model/entities_and_schema.md- entity and frontmatter model.docs/05_mcp_interface/tools_spec.md- MCP tool contracts.docs/05_mcp_interface/filesystem_tools_spec.md- filesystem tool contracts.docs/07_quality/testing_strategy.md- test strategy.docs/08_security/security_model.md- security model.docs/09_delivery/roadmap.md- milestone roadmap.CHANGELOG.md- release notes.
License
MIT.
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。