Mcpx 9aecbe05

MCP安全风控

提供方 Codestz

安全风控 未标注 Token 社区 / 第三方 Claude

mcpx是一款为MCP服务器提供安全策略、审计日志和范围化守护进程隔离的控制平面工具,适用于生产环境中的MCP服务器管理。

一句话选型结论

按资料字段整理

Mcpx 9aecbe05 属于安全风控类 MCP 服务器,传输方式未标注,Token,可直接接入 Claude,没有现成安装命令,需按官方文档接入。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类安全风控 传输方式未标注 鉴权Token 部署方式未标注
GitHub 星标 11 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

安全风控
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

安全网关Go本地部署MCP服务器管理审计日志策略执行守护进程隔离

安装与配置

未标注 Token
上游未给出传输方式
  1. 1
    准备运行环境

    按官方文档确认运行环境(常见为 Node.js 或 Python),装好后再执行安装命令。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
资产与漏洞排查

盘点资产、扫漏洞、看暴露面。

日志与审计

汇总日志、查异常行为。

合规检查

按规则检查配置与权限是否合规。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 已适配 Claude,换客户端不用重新研究配置。

先想清楚的地方

  • 没有现成安装命令,需要按官方文档自己填启动命令和参数。
  • 使用前要准备好Token对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在安全风控场景里让模型直接调用外部能力的人。
  • 已经在用 Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Mcpx 9aecbe05当前 未标注 未标注 未标注 无 2026-05-17
Cyber Strike AI 75 未标注 未标注 无 2026-05-17
mcp-for-beginners 未标注 stdio · 本地进程 本地部署 有 2026-04-13
casdoor 未标注 stdio · 本地进程 本地部署 有 2026-04-12
kubeshark 未标注 stdio · 本地进程 本地部署 有 2026-04-11
Modelcontextprotocol C22988b0 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

安全风控
字段来自项目仓库与公开索引,中文说明由本站标注
提供方Codestz上游标注的维护方 Codestz
分类安全风控本站按上游分类展示
连接方式未标注上游未给出传输方式
鉴权方式Token远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · 未标注本地安装时用到的包标识与运行时
部署类型未标注上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

10.0 KB
来自项目仓库与上游目录的原文,未做删改
                         __  __  ____ ____  __  __
                        |  \/  |/ ___|  _ \ \ \/ /
                        | |\/| | |   | |_) | \  /
                        | |  | | |___|  __/  /  \
                        |_|  |_|\____|_|    /_/\_\

          Secure gateway for MCP servers — from CLI to production.

Documentation · Examples · Installation · GitHub


mcpx is the control plane between AI agents and MCP servers. It wraps any MCP server into a CLI command with security policies, audit logging, and scoped daemon isolation — so teams can adopt MCP in production with confidence.```bash

Discover → Call → Compose

mcpx serena find_symbol --name "Auth"
mcpx serena search_for_pattern --substring_pattern "TODO" | jq -r '.file'
mcpx postgres query --sql "SELECT * FROM users LIMIT 10"


## Why mcpx

MCP servers have three problems in production:

### 1. Context cost

Loading MCP servers natively costs 50-100K tokens per session — before any work starts. mcpx calls tools on demand via Bash. Zero upfront cost.

| Setup | Context Cost | Impact |
|-------|-------------|--------|
| 5 MCP servers (native) | ~80K tokens | Context nearly full |
| **Any count (mcpx)** | **0 tokens** | **Full context available** |

### 2. No security

Every MCP tool call is unrestricted. An AI agent connected to a Postgres MCP can `DROP TABLE` as easily as `SELECT`. There's no authorization, no policy enforcement, no audit trail.

mcpx adds the missing security layer:```yaml
# Block SQL mutations on your database MCP
security:
  policies:
    - name: no-mutations
      match:
        tools: [query]
        content:
          target: args.sql
          deny_pattern: "(?i)\\b(INSERT|UPDATE|DELETE|DROP|TRUNCATE)\\b"
      action: deny
      message: "Mutation queries blocked"
``````bash
$ mcpx postgres query --sql "DROP TABLE users"
error: server "postgres": policy "no-mutations" denied tool "query"
  Reason: Mutation queries blocked

3. Multi-server management

Each server needs its own security profile. A code search MCP needs different rules than a database MCP or a messaging MCP. And when two developers work on different projects simultaneously, their daemons shouldn't interfere.

mcpx handles this with per-server security and scoped daemons:```yaml
servers:
serena:
security:
mode: editing
policies:
- name: source-only
match:
args:
relative_path: { allow_prefix: ["src/", "internal/"] }
action: deny
postgres:
security:
mode: read-only


## Installation```bash
# Homebrew (macOS / Linux)
brew tap codestz/tap
brew install mcpx

# Go install (requires Go 1.24+)
go install github.com/codestz/mcpx/cmd/mcpx@latest

# Or build from source
git clone https://github.com/codestz/mcpx.git
cd mcpx && make build

Quick Start

1. Import existing config```bash

mcpx init

Detected .mcp.json — imported 3 servers: serena, sequential-thinking, filesystem


### 2. Or configure manually```yaml
# .mcpx/config.yml
servers:
  serena:
    command: serena
    args: [start-mcp-server, --context=claude-code]
    daemon: true
    startup_timeout: 30s

3. Use it```bash

mcpx serena find_symbol --name "Auth" # call a tool
mcpx ping serena # health check
mcpx list serena -v # list all tools with flags


## Security

mcpx provides three layers of security between AI agents and MCP servers:

### Policies

Define rules that evaluate before every tool call. Supports tool name globs, argument inspection, and content regex:```yaml
security:
  enabled: true
  global:
    policies:
      - name: no-path-traversal
        match:
          args:
            "*path*": { deny_pattern: "\\.\\.\\/"}
        action: deny
        message: "Path traversal blocked"

servers:
  serena:
    security:
      mode: editing
      policies:
        - name: restrict-paths
          match:
            args:
              relative_path: { allow_prefix: ["src/", "internal/"] }
          action: deny

Modes

Quick security profiles: read-only (blocks all write tools), editing (full access), custom (policies only).```yaml
servers:
ml-pipeline:
security:
mode: read-only # prevents accidental changes to ML models


### Audit Logging

Every tool call is logged to a JSONL file with timestamps, arguments, and policy decisions:```yaml
security:
  global:
    audit:
      enabled: true
      log: "$(mcpx.project_root)/.mcpx/audit.jsonl"
      redact: ["$(secret.*)"]
``````json
{"timestamp":"2026-03-28T18:19:50Z","server":"serena","tool":"find_symbol","action":"allowed"}
{"timestamp":"2026-03-28T18:19:59Z","server":"serena","tool":"find_symbol","action":"denied","policy_name":"no-path-traversal"}

See Security Documentation for the full reference.

Configuration

Two-level config

File Scope Commit?
~/.mcpx/config.yml Global (user-level) No
.mcpx/config.yml Project-level Yes

Project config wins on conflict.

Dynamic variables```yaml

args:
- "$(mcpx.project_root)" # directory containing .mcpx/config.yml
- "$(git.root)" # git repository root
- "$(env.API_KEY)" # environment variable
- "$(secret.github_token)" # OS keychain secret
- "$(sys.os)" # linux, darwin, windows


### Server options```yaml
servers:
  my-server:
    command: string          # executable to run (required)
    args: [string]           # command arguments
    transport: stdio|http|sse # protocol (default: stdio)
    daemon: bool             # keep server alive between calls
    startup_timeout: string  # e.g. "60s" (default: "30s")
    url: string              # for http/sse transports
    headers: {}              # HTTP headers
    auth:                    # authentication
      type: bearer
      token: "$(secret.token)"
    env: {}                  # extra environment variables
    security: {}             # per-server security config

How AI Agents Use mcpx

Add this to your project's CLAUDE.md (or equivalent):```markdown

Tools

This project uses mcpx for MCP tool access:
- mcpx list — discover servers and tools
- mcpx <server> <tool> --help — get usage for any tool
- Call tools via Bash as needed


The AI discovers tools lazily, calls them on demand, and composes them through pipes.

## Secrets```bash
mcpx secret set github_token ghp_abc123    # store in OS keychain
mcpx secret list                            # list secrets
mcpx secret remove github_token             # remove

Reference in config with $(secret.) — resolved at runtime, never on disk, never logged.

Daemon Mode

Heavy servers (like Serena with LSP) benefit from staying alive:```yaml
servers:
serena:
daemon: true
``````bash
mcpx daemon status # show running daemons
mcpx daemon stop serena # stop a daemon
mcpx ping serena # health check with latency


## CLI Reference```
mcpx <server> <tool> [flags]      Call a tool
mcpx <server> <tool> --help       Show tool help
mcpx <server> <tool> --stdin      Read args from stdin JSON
mcpx <server> <tool> --example    JSON skeleton for the tool's args
mcpx <server> <tool> --validate-args ...  Type-check args without invoking
mcpx <server> --help              List all tools (compact); --full for descriptions

mcpx <server> info                Server capabilities
mcpx <server> prompt list|<name>  Prompts
mcpx <server> resource list|read  Resources

mcpx find <query>                 Rank tools across servers by intent (BM25)
mcpx batch < calls.jsonl          Run many calls in parallel (NDJSON in/out)
mcpx gain                         Token-savings dashboard (TUI)
mcpx doctor                       Config + connectivity diagnostics
mcpx ui status|stop|open|disable  Web dashboard daemon controls

mcpx list                         List servers
mcpx list <server> -v             List tools with flags
mcpx ping <server>                Health check
mcpx init                         Import .mcp.json
mcpx configure                    Generate agent reference docs (MCPX.md + per-server)

mcpx secret set|list|remove       Keychain secrets
mcpx daemon status|stop|stop-all  Daemon management
mcpx version                      Print version
mcpx completion bash|zsh|fish     Shell completions

Global flags

Flag Description
--json Output raw JSON
--quiet Suppress output
--dry-run Show what would execute
--pick Extract a JSON field
--timeout Per-call timeout

Exit codes

Code Meaning
0 Success
1 Tool error
2 Config error
3 Connection error
4 Timeout
5 Policy denied
6 Tool not found
2 Config error
3 Connection error

Architecture```

cmd/mcpx/ entrypoint
internal/
config/ YAML config loading, merging, validation
security/ policy engine, audit logging
resolver/ dynamic variable resolution: $(env.), $(git.), etc.
mcp/ MCP protocol client (JSON-RPC 2.0 over stdio/http/sse)
cli/ Cobra CLI, dynamic command generation from MCP schemas
daemon/ daemon process management, unix socket transport
secret/ OS keychain integration


- **Single binary**, zero runtime dependencies, sub-millisecond startup
- **Three transports**: stdio, HTTP (streamable), SSE
- **Security middleware**: policies evaluated before every tool call
- **Audit trail**: every call logged with secret redaction
- **No shell expansion** — `exec.Command`, not `sh -c`. No injection surface.
- **Secrets never on disk** — resolved from keychain at runtime
- **Daemon sockets** — unix socket at `/tmp/mcpx--.sock`, mode 0600

## Shell Completion```bash
echo 'eval "$(mcpx completion bash)"' >> ~/.bashrc   # Bash
echo 'eval "$(mcpx completion zsh)"' >> ~/.zshrc     # Zsh
mcpx completion fish | source                         # Fish

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT

常见问题

6 问
答案由本站依据条目字段整理,不引用上游原文
Mcpx 9aecbe05 怎么安装?

资料没有给出安装命令,请到官方仓库或文档按说明接入;这类条目多数是生态应用,或尚未发布可安装包。

Mcpx 9aecbe05 怎么在 Claude 里配置?

它的传输方式未标注,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Mcpx 9aecbe05 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Mcpx 9aecbe05 需要 API Key 吗?

资料标注需要Token认证,使用前准备好对应凭据,并按最小权限配置。

Mcpx 9aecbe05 支持哪些客户端?

资料标注支持 Claude;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Mcpx 9aecbe05 能提供哪些能力?

资料显示能力标签包括 Go、Claude、审计日志。具体工具清单以官方文档为准。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。