Context Engine MCP server
✓ VerifiedCompresses noisy logs, RAG retrieval chunks, and code context into structured, high-signal inputs optimized for LLMs.
Also known as melonelish/context-engine, context-engine, melonelish-context-engine
- Stars
- 324
- Tools
- —
- Downloads, 30 days
- 84
- Latest version
- 0.1.0
- Last commit
- Jul 11, 2026
- License
- MIT
What is Context Engine MCP?
Context Engine compresses noisy logs, RAG retrieval chunks, and code context into structured, high-signal inputs optimized for LLMs.
Summary generated from the project's README
About Context Engine MCP server
Context Engine acts as a specialized compression layer designed to transform noisy, long, or disorganized data into structured, high-signal inputs for large language models. It is built for developers working on AI agent workflows, RAG pipelines, and AI coding assistants who need to manage context limits effectively. The engine processes three specific types of data: logs, RAG retrieval chunks, and code context.
Users can interact with the system through a Python SDK, a command-line interface, or as an MCP server using the compress_context tool. It operates locally and requires Python 3.11 to 3.13. The system manages token budgets through small, medium, or large settings and enforces safety limits, such as a 200,000 character limit per text segment and a 2,000,000 byte limit for input files. By deduplicating content and prioritizing root causes or relevant evidence, it ensures that only the most critical information reaches the model.
What you can do with it
- 1Clean noisy system logsA developer can compress long tracebacks and repetitive heartbeat logs into a concise summary focusing on the root cause.
- 2Optimize RAG retrieval resultsAn AI agent can reorder retrieved document chunks to prioritize evidence that directly relates to a specific user question.
- 3Manage token budget constraintsUsers can use standardized deduplication and budget-aware cropping to fit large datasets into limited model windows.
Written from the project's README. Every use case is checked against a quote from it.
How to install Context Engine MCP
Formats verified 10 Nov 2026Install Context Engine MCP in Claude Code
$ claude mcp add context-engine -- uvx melonelish-context-engineInstall Context Engine MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"context-engine": {
"command": "uvx",
"args": [
"melonelish-context-engine"
]
}
}
}Install Context Engine MCP in VS Code
.vscode/mcp.json
{
"servers": {
"context-engine": {
"type": "stdio",
"command": "uvx",
"args": [
"melonelish-context-engine"
]
}
}
}Install Context Engine MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"context-engine": {
"command": "uvx",
"args": [
"melonelish-context-engine"
]
}
}
}Install Context Engine MCP in Codex
~/.codex/config.toml
[mcp_servers.context-engine] command = "uvx" args = ["melonelish-context-engine"]
Install Context Engine MCP in Gemini CLI
~/.gemini/settings.json
{
"mcpServers": {
"context-engine": {
"command": "uvx",
"args": [
"melonelish-context-engine"
]
}
}
}Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablemelonelish/context-engine is public and reachablegithub.com/melonelish/context-engine ↗Pass
- Package names this repository as its sourcemelonelish-context-engine names melonelish/context-engine as sourcepypi.org/project/melonelish-context-engine/ ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for melonelish-context-enginepypi.org/project/melonelish-context-engine/ ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-07-11github.com/melonelish/context-engine ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from melonelish-context-enginePass
- No critical advisory in the package or its dependencies13 resolved dependencies checked; no critical advisoryosv.dev ↗Pass
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubContext Engine - 面向 Agent / RAG / AI Coding 的上下文压缩引擎
为什么需要 Context Engine
很多 Agent / RAG / AI Coding 系统的问题,不是“没有上下文”,而是“上下文太脏、太长、太乱”。
日志里有大量重复 heartbeat 和 retry;RAG 检索结果里混着相关与不相关 chunk;代码修复场景里,模型经常拿到一堆文件,却缺少最小修复线索。
Context Engine 解决的是这个中间层问题:
在把内容交给大模型之前,先按任务类型压缩、排序、去噪,并输出结构化的高信号上下文。
它不是普通摘要工具,而是一个更偏工程化的上下文整理层。
能力边界问题
| 场景 | 直接丢给模型 | Context Engine |
|---|---|---|
| 长日志 / traceback | 复制全部日志,或者简单截断 | 保留疑似根因、traceback 尾部,折叠重复噪声 |
| RAG 检索结果 | 按检索顺序塞入所有 chunk | 围绕用户问题重新排序,区分高信号与低信号证据 |
| AI Coding 修复 | 把附近文件全部塞进去 | 根据 issue / test output 排序 hotspot file |
| 输入异常 | 抛原始 traceback | 返回结构化错误、错误码和修复提示 |
| Token 不够 | 从头或从尾硬切 | 在标准化和去重后按预算裁剪 |
系统架构
核心理念:先把上下文变成可解释、可预算、可消费的结构,再交给 LLM。
flowchart TB
classDef input fill:#fff7d6,stroke:#d99b00,stroke-width:2px,color:#5c3b00;
classDef core fill:#e7f1ff,stroke:#2563eb,stroke-width:2px,color:#123c8c;
classDef mode fill:#eaf8ef,stroke:#1f8a4c,stroke-width:2px,color:#0f4f2b;
classDef output fill:#f5ecff,stroke:#7c3aed,stroke-width:2px,color:#3f1d7a;
subgraph Inputs["输入层"]
CLI["CLI<br/>本地文件 / 示例"]:::input
SDK["Python API<br/>CompressionRequest"]:::input
MCP["MCP Tool<br/>compress_context"]:::input
end
subgraph Core["核心 Pipeline"]
Validate["输入校验<br/>schema + size guard"]:::core
Normalize["标准化<br/>ContextItem list"]:::core
Dedupe["去重<br/>重复内容折叠"]:::core
Budget["预算控制<br/>small / medium / large"]:::core
end
subgraph Compressors["任务感知压缩器"]
Logs["logs<br/>root cause + traceback"]:::mode
Rag["rag<br/>question-aware evidence"]:::mode
Code["code<br/>hotspot files + failure signal"]:::mode
end
subgraph Outputs["输出契约"]
Result["ok: true<br/>summary + key facts + llm_ready_context"]:::output
Error["ok: false<br/>error_code + hint + details"]:::output
end
CLI --> Validate
SDK --> Validate
MCP --> Validate
Validate --> Normalize --> Dedupe --> Budget
Budget --> Logs
Budget --> Rag
Budget --> Code
Logs --> Result
Rag --> Result
Code --> Result
Validate --> Error
数据流
sequenceDiagram
autonumber
participant U as 用户 / Agent
participant C as CLI / MCP / SDK
participant V as Validator
participant P as Pipeline
participant M as Mode Compressor
participant O as Output Envelope
U->>C: 提交 logs / rag / code 输入
C->>V: 校验字段、大小和模式
V->>P: 标准化为 ContextItem
P->>P: 去重、排序、预算裁剪
P->>M: 进入任务压缩器
M->>O: 生成 summary / key_facts / llm_ready_context
O-->>U: 返回结构化结果或结构化错误
核心能力
1. logs:根因导向的日志压缩
logs 模式面向长日志、重复日志、异常栈和混合噪声。
| 能力 | 说明 |
|---|---|
| 根因提取 | 优先保留 error、exception、failed、fatal 等关键行 |
| Traceback 保留 | 保留更接近根因的 traceback 尾部 |
| 噪声折叠 | 把重复 heartbeat、poll、retry、duplicate line 归并成计数 |
| LLM-ready 输出 | 生成可以直接放进诊断或修复 prompt 的上下文块 |
2. rag:围绕问题的证据重排
rag 模式接收 question 和 chunks,把检索结果从“按召回顺序堆叠”变成“围绕问题排序”。
| 分层 | 含义 |
|---|---|
HIGH |
与问题重叠度高,可能直接支持回答 |
SUPPORT |
有帮助但不是核心证据 |
| Low signal | 相关性弱,不应该占据主要上下文窗口 |
3. code:最小修复上下文
code 模式接收 issue、可选 test_output 和 files,用于给 AI Coding 助手准备更聚焦的修复输入。
| 信号 | 作用 |
|---|---|
| Issue terms | 保留用户描述的问题焦点 |
| Failure terms | 把测试失败、异常信息和候选文件关联起来 |
| Hot path hints | 对 test、parser、pipeline、service、validator 等路径加权 |
| Supporting files | 保留辅助上下文,但不让它淹没主线 |
4. CLI + MCP 双入口
同一套压缩逻辑可通过三种方式使用:
| 入口 | 用途 |
|---|---|
| Python API | 集成到自己的包或服务里 |
| CLI | 本地处理日志、样例和脚本任务 |
| MCP Server | 接入支持 MCP 的 Agent / IDE / 自动化环境 |
快速开始
前置要求
| 项目 | 要求 |
|---|---|
| Python | 3.11 到 3.13 |
| 包管理器 | pip |
| MCP 运行时 | 可选,通过 mcp extra 安装 |
安装
python -m venv .venv
.venv\Scripts\Activate.ps1
python -m pip install -e .[dev,mcp]
从 PyPI 安装发布包:
python -m pip install "melonelish-context-engine[mcp]"
运行示例
python -m pytest -q
context-engine --mode logs --input examples/logs/sample.log --budget medium
context-engine --mode rag --input examples/rag/sample.json --budget medium
context-engine --mode code --input examples/code/sample.json --budget medium
CLI 使用
context-engine --mode logs --input examples/logs/sample.log --budget small
context-engine --mode rag --input examples/rag/sample.json --budget medium
context-engine --mode code --input examples/code/sample.json --budget large
输入格式
logs:纯文本
2026-07-03 10:01:12 ERROR payment.worker failed to charge order
Traceback (most recent call last):
...
ValueError: missing customer_id
rag:JSON
{
"question": "Why did checkout fail?",
"chunks": [
{
"content": "Checkout fails when customer_id is missing.",
"metadata": {"source": "runbook.md"}
}
]
}
code:JSON
{
"issue": "Checkout test fails when customer_id is omitted.",
"test_output": "ValueError: missing customer_id",
"files": [
{
"path": "src/payments/checkout.py",
"content": "def checkout(order): ..."
}
]
}
MCP 使用
通过 stdio 启动 MCP Server:
context-engine-mcp
当前暴露一个工具:
| 工具 | 参数 | 功能 |
|---|---|---|
compress_context |
mode、budget、content 或 payload |
压缩 logs / rag / code 上下文 |
示例错误返回:
{
"ok": false,
"error": {
"error_code": "invalid_field",
"message": "Field 'chunks' must be a non-empty list.",
"hint": "Provide at least one item in 'chunks'."
}
}
输出契约
CLI 和 MCP 都返回统一结构:
| 结果 | 结构 |
|---|---|
| 成功 | { "ok": true, "result": ... } |
| 失败 | { "ok": false, "error": { "error_code": "...", "message": "...", "hint": "...", "details": ... } } |
成功结果会包含 schema version、summary、key facts、被丢弃或降权的噪声,以及 llm_ready_context。
安全边界
当前内置限制:
| 限制项 | 当前值 |
|---|---|
| 单段文本最大长度 | 200000 字符 |
| 结构化列表最大数量 | 64 项 |
| 输入文件最大大小 | 2000000 bytes |
| Schema version | 1.0 |
这些限制用于避免外部工作流把任意超大 payload 直接打进工具。
当前已支持
- Python
3.11到3.13 logs/rag/code三种模式- 三种模式的 CLI 使用
- 通过
compress_context暴露 MCP 工具 - 纯文本日志输入
- JSON 格式的 RAG 和 code 输入
- 结构化成功 / 失败 envelope
- 基础 benchmark 和 GitHub Actions CI
暂不支持
- PDF、图片、Office 文件等二进制输入
- embedding-aware reranking
- 仓库级依赖图分析
- 跨未来大版本的长期兼容性承诺
- 生产级鉴权、持久化、监控和审计能力
Benchmark
Benchmark 资料位于:
当前样例集体现的行为:
| 模式 | 对比对象 | 当前效果 |
|---|---|---|
logs |
原始日志 / 简单截断 | 保留根因,折叠重复噪声 |
rag |
直接 dump 检索 chunk | 围绕问题排序证据 |
code |
普通文件摘要 | 保留 issue、失败信号、hotspot file 和支持上下文 |
v0.1.0 的 benchmark 还很小,适合作为回归检查和展示样例,不代表完整生产评测。
开发
python -m pip install -e .[dev,mcp]
python -m pytest -q
python benchmarks/generate_benchmarks.py
CI 会在 push 和 pull request 时运行安装、测试和 benchmark 生成检查。
发布状态
当前版本目标:v0.1.0
已发布渠道:
- PyPI: https://pypi.org/project/melonelish-context-engine/
- MCP Registry: https://registry.modelcontextprotocol.io/v0/servers/io.github.melonelish/context-engine
这个版本适合早期外部试用、集成测试和开发者工作流验证。它已经具备可复用包结构、测试、文档,以及可直接安装的 PyPI / MCP Registry 发布入口,但仍然是范围明确的早期 beta。
Roadmap
- 引入 embedding 或 reranker 驱动的 RAG 排序。
- 加强 code hotspot 识别,加入更可靠的结构化代码信号。
- 扩展 benchmark,加入更多真实脏数据样本。
- 发布更易安装的正式包版本。
- 增加更多 Agent Runtime / MCP 集成示例。
License
本项目采用 MIT License。
Context Engine MCP questions
How do I add Context Engine MCP to Claude Code?
Run claude mcp add context-engine -- uvx melonelish-context-engine in your terminal. Add --scope user to make it available in every project.
How do I add Context Engine MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
What modes does Context Engine support?
It supports three specialized modes: logs for root cause extraction, rag for question-aware evidence reordering, and code for minimal repair context.
How can I use Context Engine with an MCP client?
You can use it as an MCP Server by running the command context-engine-mcp, which exposes the compress_context tool.
What are the input limitations for security?
The engine has built-in limits including a maximum single text length of 200,000 characters, a maximum of 64 structured items, and a maximum input file size of 2,000,000 bytes.