JN Engine Contributor MCP
A read-only MCP gateway for contributors to JN Engine, exposing immutable repository snapshots via REST and Streamable HTTP for ChatGPT, Claude, and coding agents.
README
JN Engine Contributor MCP
A bounded REST and Streamable HTTP MCP gateway for contributors working on JN Engine. It exposes a reviewed, immutable repository snapshot to ChatGPT, Claude, MCP-compatible coding agents, and ordinary HTTP clients without giving those clients shell or Git access to the host.
This is the public, contributor-safe distribution. It contains the complete service, tests, snapshot exporters, Docker configuration, and example deployment files. It does not contain production credentials, personal paths, private hostnames, audit records, incident reports, or mutable operator data.
What It Provides
The engine profile exposes nine tools:
searchandfetchfor commit-bound source and documentation retrieval;list_tasksandproject_contextfor bounded project orientation;claim_taskandrelease_taskfor audited, expiring task ownership;lookup_symbolfor recovered symbol/address lookup; andcheck_statusfor fail-closedcoreandassetsGitHub Actions status;open_prfor a narrowly constrained contributor pull-request path.
An optional source profile exposes this repository itself through search, fetch,
and repository_context. That profile lets a web or app-based agent inspect the MCP
implementation even when the platform's account-level GitHub integration is not
available as a callable chat tool.
Both profiles use immutable snapshots. Search IDs include the active commit and become invalid after a snapshot refresh, so clients must search again rather than silently mixing content from different revisions.
Why Sanitization Does Not Reduce Contributor Value
The useful parts are present: application behavior, interfaces, tests, fixture data, dependency locks, container boundaries, deployment examples, and contribution rules. Only environment-specific material was excluded. Contributors can:
- run all unit and integration tests;
- build and inspect the container;
- create snapshots from the public JN Engine repository;
- exercise either MCP profile locally without authentication;
- configure their own GitHub OAuth deployment;
- develop new bounded tools and submit pull requests; and
- use the source profile to ground ChatGPT or Claude in this implementation.
See Intended Usage for contributor and operator workflows, and Public Repository Boundary for the exact public/private split.
Quick Start
Requirements: Python 3.11 through 3.13, Git, and ripgrep.
python3 -m venv .venv
.venv/bin/python -m pip install --constraint requirements.lock -e '.[test]'
Build a local immutable snapshot of JN Engine:
JN_REPO_MIRROR_PATH="$PWD/.local/jn-engine.git" \
JN_SNAPSHOT_ROOT="$PWD/.local/snapshots" \
deploy/refresh_snapshot.sh --build-only
The command prints the promoted snapshot path. Start a loopback-only development server with that path:
APP_ENV=development \
AUTH_MODE=authless_local \
API_HOST=127.0.0.1 \
JN_SNAPSHOT_PATH="$PWD/.local/snapshots/<commit>" \
.venv/bin/python -m app.run
Connect an MCP client to http://127.0.0.1:8788/mcp. Authless mode is deliberately
restricted to a loopback listener and is not a production configuration.
Verification
CI constructs its own immutable JN Engine fixture before running the suite. Locally,
set JN_TEST_SNAPSHOT_PATH to the snapshot created above:
JN_TEST_SNAPSHOT_PATH="$PWD/.local/snapshots/<commit>" \
.venv/bin/python -m pytest -q
Production GitHub OAuth configuration and secret-file permissions are documented in Deployment. MCP contracts are in MCP Surface.
Security Properties
- Repository identity and ref are code-owned and cannot be redirected by environment variables.
- Snapshot files are inventory checked, contained, non-symlink, regular, bounded, and read-only in production.
- OAuth state is encrypted before being stored in a mode-0700 file-tree directory.
- Secret values are accepted through mode-0600 files, not command-line arguments or URLs.
- Expiring task ownership uses a dedicated versioned mode-0600 ledger, separate from the rotatable status/PR audit log.
- Write actions require the authenticated engine profile and a dedicated PR token; shell actions always fail startup when enabled.
- The container runs non-root with a read-only root filesystem, no capabilities, no Docker socket, and no Git executable.
Report vulnerabilities privately as described in SECURITY.md.
License
Licensed under the MIT License. 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。