Paulbellamy MCP

MCP开发工具

提供方 paulbellamy

开发工具 未标注 Token 社区 / 第三方 ClaudeCursor

一个用于发现和调用外部MCP服务器上工具的CLI工具,适用于需要灵活工具调用和跨服务器搜索的场景。

一句话选型结论

按资料字段整理

Paulbellamy MCP 属于开发工具类 MCP 服务器,传输方式未标注,Token,可直接接入 Claude、Cursor,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。

分类开发工具 传输方式未标注 鉴权Token 部署方式未标注
GitHub 星标 0 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

开发工具
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

CLI工具Go本地部署MCP协议工具发现跨服务器搜索开发效率

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
代码与仓库操作

在对话里直接查代码、提 issue、走 PR,省掉来回切工具。

本地命令与脚本

把构建、测试、格式化这类重复命令交给模型执行。

接口联调与测试

让模型调你的接口、跑一遍用例,快速定位问题。

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、Cursor,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

  • 需要在开发工具场景里让模型直接调用外部能力的人。
  • 已经在用 Claude、Cursor 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
Paulbellamy MCP当前 未标注 未标注 未标注 无 2026-05-17
Git 838e2ec1 12 stdio · 本地进程 本地部署 有 2026-05-17
Markitdown MCP 2c653a5f 1 stdio · 本地进程 本地部署 有 2026-05-17
JavaGuide 未标注 stdio · 本地进程 本地部署 有 2026-04-11
awesome-claude-skills 未标注 stdio · 本地进程 本地部署 有 2026-04-26
Context7 831acf40 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

开发工具
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paulbellamy上游标注的维护方 paulbellamy
分类开发工具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

6.8 KB
来自项目仓库与上游目录的原文,未做删改

mcp

A CLI tool for discovering and calling tools on external MCP (Model Context Protocol) servers.

When to use this CLI

Most agent harnesses (Claude Code, Cursor, etc.) can connect to MCP servers
natively — every tool on every connected server is loaded into the prompt as a
typed function the model can call directly. That works well when you need a
small, fixed set of tools every turn. It works badly when the catalogue is
large, the server is one-off, or you want to compose tool output with the
shell.

This CLI is the second path. Reach for it when native loading is wrong for the
job.

Decision rule (one sentence): if the tool you need is already loaded as
mcp____ in your tool list, call it natively; otherwise use
this CLI.

Use this CLI when:
- The server isn't loaded natively (no harness config, ad-hoc URL, dev sandbox).
- The catalogue is large but you only need one or two tools this session — pay
for one schema, not all of them.
- You need to search across every connected server at once (mcp tools --query).
- You want to pipe tool output through jq, capture it in a variable, or
script it.
- You need to see what each server costs in tokens (mcp stats).
- Your agent has no native MCP support and shell is the only channel.

Use native MCP when:
- The tool is already loaded and you'll call it many times this session — the
schema-load cost amortizes.
- Args are deeply nested or typed — the model emits them directly, no JSON
escaping.
- You're inside a tight loop where the extra mcp tools / mcp schema
round-trips matter.

What this CLI gives you that native loading doesn't:
- Lazy schema loading. mcp tools returns name+description only. Schemas
(often 10k+ tokens per server) are fetched one tool at a time, on demand,
via mcp schema. Native loading pays the full schema tax every turn whether
you call the tool or not.
- Ad-hoc URLs. mcp tools https://api.example.com/mcp works without
registering the server. Same for call, ping. Token auth via
MCP_AUTH_TOKEN.
- Cross-server search. mcp tools --query foo ranks tools by keyword
across every connected server. Native has no equivalent.
- Resource access. mcp resources lists a server's readable resources and
templates; mcp read fetches their contents. Useful for servers (like
Notion) that expose data as resources rather than tools.
- Token visibility. mcp stats (and --full) tells you exactly what
each server's schemas would cost if loaded natively.
- Streaming. mcp call --stream emits NDJSON progress events for
long-running tools.
- Shell composability. Output is JSON on stdout, logs on stderr, exit
codes for success/failure. Pipe, capture, chain at will.

Tradeoffs to know:
- More turns per call (list → schema → call) than a native one-shot invocation.
- Args are passed as a JSON string — escaping matters.
- Schema cache can drift; use --refresh if a server changed.
- Stdio servers have a cold-start cost on first call (use mcp daemon to keep
them warm).

Install

Claude Code plugin```bash

claude plugin marketplace add https://github.com/paulbellamy/mcp
claude plugin install mcp-cli


### Pi

Pi ships without built-in MCP support, so this CLI is a
natural fit. The repo is a Pi package:
installing it registers the `mcp-cli` skill and an extension that downloads the
`mcp` binary on first session start.```bash
pi install git:github.com/paulbellamy/mcp

Binary```bash

curl -fsSL https://raw.githubusercontent.com/paulbellamy/mcp/master/scripts/install.sh | sh


### From source```bash
go install github.com/paulbellamy/mcp@latest

Usage```bash

Overview: one line per server with its tool names (or why it failed)

mcp list [--refresh]

List configured servers (full config as JSON)

mcp servers

Add a server (HTTP or stdio)

mcp add <name> <url>
mcp add <name> --stdio <command> [args...]

Add an HTTP server with static headers (repeatable; --header or -H).

Header values may reference environment variables as ${VAR}, expanded at

request time so secrets never touch servers.json. This is how you connect

to enterprise MCP servers that authenticate with a static key + org id

rather than OAuth (e.g. Devin):

mcp add devin https://mcp.devin.ai/mcp \
--header "Authorization: Bearer ${DEVIN_API_KEY}" \
--header "X-Org-Id: ${DEVIN_ORG_ID}"

Discover tools (compact summaries — name + description only)

mcp tools [server] [--query "search term"] [--refresh]

Include full inputSchema for every tool (large; usually not needed)

mcp tools --full

Fetch the full schema for one tool on demand

mcp schema <server> <tool>

List readable resources (and resource templates) — name + uri + description

mcp resources [server] [--query "search term"]

Read a resource's contents by URI

mcp read <server> <uri>

Estimate the token cost of cached schemas across servers

mcp stats [--full]

Call a tool

mcp call <server> <tool> --params '{"key": "value"}'

Scalar params can also be passed as individual flags (coerced via the cached

schema). Array/object params can't be a bare string — pass inline JSON for the

flag, or use --params:

mcp call <server> <tool> --query hello --limit 10
mcp call <server> <tool> --cmd '["bash","-lc","cd /app && npm run build"]'

Stream progress for long-running tools

mcp call <server> <tool> --stream --params '{"key": "value"}'

Override the per-call timeout (default: 2m plain / 5m streaming; 0 = no limit)

mcp call <server> <tool> --timeout 10m --params '{"key": "value"}'

Authenticate

mcp auth <name> --callback-url <url>

Use a server without adding it (pass URL directly)

mcp tools https://api.example.com/mcp
mcp call https://api.example.com/mcp <tool> --params '{"key": "value"}'
mcp resources https://api.example.com/mcp
mcp read https://api.example.com/mcp <uri>
mcp ping https://api.example.com/mcp

Authenticate with a token for ad-hoc URLs

MCP_AUTH_TOKEN=<token> mcp call https://api.example.com/mcp <tool> --params '{}'

Static headers for ad-hoc URLs (no add needed): one "Name: Value" per

line in MCP_HEADERS. Values may reference env vars as ${VAR}. Works for

every ad-hoc command (tools, call, resources, read, ping, listen).

MCP_HEADERS="Authorization: Bearer ${DEVIN_API_KEY}
X-Org-Id: ${DEVIN_ORG_ID}" mcp tools https://mcp.devin.ai/mcp

Ping / remove

mcp ping <server>
mcp remove <name>

Progress chatter (token refresh, schema-refresh retries, skipped transport

lines) is off by default so stderr carries only warnings and errors.

Turn it on with a global flag before the subcommand, or via env.

mcp --verbose call <server> <tool>
MCP_VERBOSE=1 mcp call <server> <tool>

Print the agent skill (SKILL.md) bundled with this binary. Useful for agents

without a plugin/skill system: pipe it into their instructions.

mcp --skill
```

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
Paulbellamy MCP 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。

Paulbellamy MCP 怎么在 Claude、Cursor 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。

Paulbellamy MCP 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。

Paulbellamy MCP 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。

Paulbellamy MCP 支持哪些客户端?

资料标注支持 Claude、Cursor;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。

Paulbellamy MCP 能提供哪些能力?

资料显示能力标签包括 命令行工具、工具发现、Go、Claude、开发工具。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。