能力标签
UTCP工具
🔌
MCP工具

UTCP工具

基于 Go · 让 AI 助手直接操作你的系统与工具
英文名:go-utcp
⭐ 119 Stars 🍴 9 Forks 💻 Go 📄 MPL-2.0 🏷 AI 8.0分
8.0AI 综合评分
mcpaigolang
✦ AI Skill Hub 推荐

AI Skill Hub 强烈推荐:UTCP工具 是一款优质的MCP工具。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的MCP工具解决方案,这是一个值得深入了解的选择。

📚 深度解析

UTCP工具 是一款基于 MCP(Model Context Protocol)标准协议的 AI 工具扩展。MCP 协议由 Anthropic 开发并开源,旨在建立 AI 模型与外部工具之间的标准化通信接口,目前已被 Claude Desktop、Claude Code、Cursor 等主流 AI 工具采纳。

通过安装 UTCP工具,你的 AI 助手将获得额外的工具调用能力,可以用自然语言直接操控该工具的功能,无需学习复杂的命令行语法。MCP 工具的核心价值在于"一次配置,永久增强"——配置完成后,每次与 AI 对话时都可以无缝调用这些工具。

在技术实现上,MCP 工具通过标准的 JSON-RPC 协议与 AI 客户端通信,工具的功能以"工具列表"的形式暴露给 AI 模型,AI 可以按需调用。UTCP工具 提供了结构化的工具调用接口,使 AI 模型能够精确地理解和使用每个功能点,显著降低 AI 在工具使用上的错误率。

与传统的 API 集成相比,MCP 工具的优势在于无需编写代码——用户只需在配置文件中添加几行 JSON,即可让 AI 获得全新能力。AI Skill Hub 将 UTCP工具 评为 AI 评分 8.0 分,属于同类工具中的优质选择。

📋 工具概览

UTCP工具 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。

GitHub Stars
⭐ 119
开发语言
Go
支持平台
Windows / macOS / Linux(跨平台)
维护状态
轻量级项目,按需更新
开源协议
MPL-2.0
AI 综合评分
8.0 分
工具类型
MCP工具
Forks
9

📖 中文文档

以下内容由 AI Skill Hub 根据项目信息自动整理,如需查看完整原始文档请访问底部「原始来源」。

UTCP工具 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。

📌 核心特色
  • 通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
  • 提供结构化工具调用接口,显著降低 AI 集成复杂度
  • 支持 Claude Desktop 和 Claude Code 无缝接入,开箱即用
  • 可与其他 MCP 工具组合叠加,构建完整 AI 工作站
  • 轻量无侵入设计,不影响现有系统架构
🎯 主要使用场景
  • 在 Claude Desktop 对话中直接调用本地工具,实现 AI 与系统的深度联动
  • 通过自然语言驱动复杂的多步骤自动化任务,代替繁琐手动操作
  • 将多个 MCP 工具组合使用,构建个人专属 AI 工作站
以下安装命令基于项目开发语言和类型自动生成,实际以官方 README 为准。
安装命令
# 方式一:通过 Claude Code CLI 一键安装
claude skill install https://github.com/universal-tool-calling-protocol/go-utcp

# 方式二:手动配置 claude_desktop_config.json
{
  "mcpServers": {
    "utcp--": {
      "command": "npx",
      "args": ["-y", "go-utcp"]
    }
  }
}

# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json
📋 安装步骤说明
  1. 确认已安装 Node.js(v18 或以上版本)
  2. 打开 Claude Desktop 或 Claude Code 的 MCP 配置文件
  3. 按「交给 Agent 安装 → Claude Desktop」标签中的 JSON 配置填入 mcpServers 字段
  4. 保存配置文件并重启 Claude 客户端
  5. 重启后,在对话中即可使用本工具
以下用法示例由 AI Skill Hub 整理,涵盖最常见的使用场景。
常用命令 / 代码示例
# 安装后在 Claude 对话中直接使用
# 示例:
用户: 请帮我用 UTCP工具 执行以下任务...
Claude: [自动调用 UTCP工具 MCP 工具处理请求]

# 查看可用工具列表
# 在 Claude 中输入:"列出所有可用的 MCP 工具"
以下配置示例基于典型使用场景生成,具体参数请参照官方文档调整。
配置示例
// claude_desktop_config.json 配置示例
{
  "mcpServers": {
    "utcp__": {
      "command": "npx",
      "args": ["-y", "go-utcp"],
      "env": {
        // "API_KEY": "your-api-key-here"
      }
    }
  }
}

// 保存后重启 Claude Desktop 生效
📑 README 深度解析 真实文档 完整度 95/100 查看 GitHub 原文 →
以下内容由系统直接从 GitHub README 解析整理,保留代码块、表格与列表结构。

go-utcp

MCP vs. UTCP

Go Report Card Go Reference License: MPL 2.0

go-utcp is the Go implementation of the Universal Tool Calling Protocol (UTCP). It provides a single client-side abstraction for discovering, describing, searching, invoking, and streaming tools exposed through heterogeneous providers and transports.

The project is designed for applications that need to connect an agent, service, workflow engine, or automation runtime to many kinds of tools without coupling application logic to every transport individually.

One abstraction, many transports

Application code should not have to understand every transport implementation.

Features

  • Unified discovery API.
  • Unified tool invocation API.
  • Streaming tool invocation.
  • Provider registration and deregistration.
  • Provider-scoped tool names.
  • In-memory tool repository.
  • Pluggable repository and search abstractions.
  • HTTP and OpenAPI support.
  • CLI process providers.
  • Server-Sent Events.
  • Streamable HTTP.
  • WebSocket.
  • gRPC and gNMI support.
  • GraphQL queries and subscriptions.
  • TCP and UDP transports.
  • WebRTC data-channel transport.
  • MCP over supported MCP transports.
  • Local text-template providers.
  • Environment variable substitution.
  • .env loading.
  • Runtime variables.
  • CodeMode tool composition.
  • Streaming result handling.
  • Context-aware calls.
  • Standalone examples.
  • Go-native APIs.

Requirements

  • Go 1.25 or newer.
  • Network access for network-backed providers.
  • Provider-specific runtime dependencies where applicable.

Check the module and examples for transport-specific requirements before deploying a particular provider.

Installation

Install the library with:

go get github.com/universal-tool-calling-protocol/go-utcp@latest

Then import it from Go:

import utcp "github.com/universal-tool-calling-protocol/go-utcp"

CodeMode is available under its plugin package:

import "github.com/universal-tool-calling-protocol/go-utcp/src/plugins/codemode"

Quick start

The smallest useful example can use a local text provider, so no external server is required.

Create providers.json:

{
  "providers": [
    {
      "provider_type": "text",
      "name": "greetings",
      "templates": {
        "hello": "Hello, {{.name}}!"
      }
    }
  ]
}

Create a Go program:

package main

import (
    "context"
    "fmt"
    "log"

    utcp "github.com/universal-tool-calling-protocol/go-utcp"
)

func main() {
    ctx := context.Background()

    client, err := utcp.NewUTCPClient(ctx, &utcp.UtcpClientConfig{
        ProvidersFilePath: "providers.json",
    }, nil, nil)
    if err != nil {
        log.Fatal(err)
    }

    tools, err := client.SearchTools("", 10)
    if err != nil {
        log.Fatal(err)
    }

    for _, tool := range tools {
        fmt.Printf("%s: %s\n", tool.Name, tool.Description)
    }

    result, err := client.CallTool(ctx, "greetings.hello", map[string]any{
        "name": "UTCP",
    })
    if err != nil {
        log.Fatal(err)
    }

    fmt.Println(result)
}

Run it with:

go run .

The important detail is the qualified tool name:

<provider>.<tool>

For the example above:

greetings.hello

Provider qualification prevents ambiguous tool names when several providers expose tools with the same local name.

Basic CodeMode example

import "github.com/universal-tool-calling-protocol/go-utcp/src/plugins/codemode"

cm := codemode.NewCodeModeUTCP(client, nil)

result, err := cm.Execute(ctx, codemode.CodeModeArgs{
    Code: `
        value, err := codemode.CallTool("greetings.hello", map[string]any{
            "name": "CodeMode",
        })
        if err != nil {
            __out = err
            return
        }
        __out = value
    `,
    Timeout: 5_000,
})

The exact sandbox restrictions and supported language surface are implementation details and should be checked against the current CodeMode package.

Examples

The examples directory contains standalone examples for providers and transports.

A typical example can be run from its own directory:

cd examples/text_client
GOWORK=off go run -mod=mod .

The GOWORK=off setting is useful when an example is maintained as its own Go module and should use its own dependency graph.

Example categories

Examples cover patterns such as:

  • Basic client creation.
  • Provider configuration.
  • Text tools.
  • HTTP providers.
  • Streaming.
  • MCP integrations.
  • CodeMode.
  • Transport-specific clients.

Always inspect the example's local README or source before running a network example because some examples expect a local service to be running.

Performance guidelines

  • Reuse a client where appropriate.
  • Avoid repeatedly discovering the same provider unnecessarily.
  • Keep provider metadata scoped to what the application needs.
  • Use streaming when incremental results are valuable.
  • Apply context deadlines.
  • Avoid unnecessary serialization between agent layers.
  • Measure the real provider path rather than optimizing only local code.

Repository examples

The examples directory is intentionally part of the documentation surface. When behavior changes, examples should be updated if their assumptions no longer match the implementation.

A good example should:

  • Be small.
  • Be runnable.
  • Show realistic configuration.
  • Avoid unnecessary dependencies.
  • Explain external prerequisites.
  • Demonstrate correct error handling.

CodeMode

CodeMode lets a constrained Go-like program compose multiple tool calls in one execution. This is useful when an agent needs loops, branching, transformation, or multi-step orchestration.

CodeMode

CodeMode provides a controlled execution environment for composing multiple registered tool calls.

The motivation is simple: an agent sometimes needs more than one independent tool invocation. It may need to:

  1. Search for resources.
  2. Iterate over results.
  3. Call another tool for each item.
  4. Transform intermediate values.
  5. Branch based on results.
  6. Return one final value.

Doing every intermediate step through separate model round trips can be expensive. CodeMode moves some orchestration into a constrained program.

CodeMode helpers

CodeMode snippets can use helpers including:

  • codemode.CallTool.
  • codemode.CallToolStream.
  • codemode.SearchTools.

The resulting CodeModeResult contains the produced value and captured output streams.

Why CodeMode?

Without CodeMode, an agent may need to repeatedly request a tool, wait for a result, send the result back to the model, and request the next tool. CodeMode can reduce this orchestration overhead for deterministic multi-step logic.

Tool safety in CodeMode

CodeMode should not be treated as unrestricted host execution. Tool access should remain bounded by the registered provider and the CodeMode execution policy.

Applications embedding CodeMode should carefully define:

  • Which tools are available.
  • Which providers are trusted.
  • Execution timeouts.
  • Resource limits.
  • Filesystem access.
  • Network access.
  • Process access.
  • Error handling.

Is CodeMode required?

No. CodeMode is an optional plugin for applications that need programmatic multi-tool composition.

Treating CodeMode as unrestricted execution

CodeMode should have explicit tool and execution boundaries.

Provider configuration

ProvidersFilePath can point at a JSON document. The loader accepts several root shapes.

Configuration management

For small applications, a single providers.json file can be sufficient.

For larger deployments, consider generating or assembling provider configuration from:

  • Environment-specific templates.
  • Secret managers.
  • Deployment configuration.
  • Service discovery.
  • Tenant configuration.

Keep the distinction between configuration and credentials clear.

A provider definition should describe how a capability is reached. A secret manager should provide sensitive values needed to authenticate to it.

Environment variable is not resolved

Check:

  • Variable spelling.
  • $NAME versus ${NAME} syntax.
  • Variables configuration.
  • .env path.
  • Process environment.
  • Configuration load order.

Do not print secret values while debugging.

Client API

MethodPurpose
RegisterToolProviderDiscover and store tools from a provider.
DeregisterToolProviderRemove a provider and its tools.
SearchToolsSearch registered tools.
CallToolInvoke a tool synchronously.
CallToolStreamInvoke a streaming tool.
GetTransportsAccess registered transport implementations.

The Go API may grow as UTCP evolves. For the authoritative signatures and types, use the package documentation on pkg.go.dev and the source tree.

OpenAPI

The HTTP integration can discover tools from OpenAPI-backed services.

OpenAPI is useful when an existing API already describes operations, parameters, request bodies, and responses through a machine-readable specification.

A practical integration flow is:

OpenAPI document
      |
      v
HTTP provider
      |
      v
UTCP discovery
      |
      v
Tool metadata
      |
      v
Agent / application

When using OpenAPI discovery, keep the generated tool surface intentionally scoped. Large APIs can expose hundreds of operations, and applications may benefit from selecting only the operations required by a particular workflow.

CLI provider fails

Check:

  • Executable availability.
  • PATH.
  • File permissions.
  • Working directory.
  • Environment variables.
  • Argument encoding.
  • Process exit status.

For production workloads, prefer narrowly defined commands over arbitrary shell execution.

Minimal reference application

A compact application generally needs only four steps:

ctx := context.Background()

client, err := utcp.NewUTCPClient(ctx, config, nil, nil)
if err != nil {
    return err
}

tools, err := client.SearchTools("", 10)
if err != nil {
    return err
}

_ = tools

result, err := client.CallTool(ctx, "provider.tool", map[string]any{})
if err != nil {
    return err
}

_ = result

This is the core mental model of go-utcp: configure providers, discover tools, select a tool, and call it.

Integration tests

Network-backed integrations should be tested with deterministic local servers where possible.

Tests should cover:

  • Successful discovery.
  • Successful invocation.
  • Invalid configuration.
  • Missing tools.
  • Provider deregistration.
  • Context cancellation.
  • Stream completion.
  • Stream failure.
  • Authentication failures.
  • Malformed provider responses.

Go modules

Use the repository's declared Go version and module dependencies. Avoid introducing unnecessary dependencies for functionality that can be implemented with the standard library or existing project abstractions.

Agent integration

go-utcp can act as the tool layer underneath an agent runtime.

A common architecture is:

                   User
                    |
                    v
                 Agent
                    |
              Tool selection
                    |
                    v
               go-utcp
                    |
       +------------+-------------+
       |            |             |
      HTTP         MCP          gRPC
       |            |             |
       v            v             v
    Service      Server        Service

The agent can search available tools rather than hard-coding every integration.

Go-native integration

The library should feel natural in Go applications, using contexts, errors, interfaces, maps, and standard tooling.

Troubleshooting

FAQ

Security review questions

Before exposing a tool to an autonomous caller, ask:

  1. What can this tool change?
  2. What credentials does it use?
  3. What data can it read?
  4. Can the operation be reversed?
  5. Is it idempotent?
  6. What is the blast radius of misuse?
  7. Does it need human approval?
  8. Can arguments be constrained?
  9. Can output contain secrets?
  10. What happens if the provider is compromised?

These questions are especially important for filesystem, shell, database, deployment, billing, and administrative tools.

🎯 aiskill88 AI 点评 A 级 2026-07-13

高质量的MCP工具,Go语言实现

⚡ 核心功能

👥 适合人群

Claude Desktop / Claude Code 用户AI 工具开发者需要扩展 AI 能力的专业人士自动化工程师

🎯 使用场景

  • 在 Claude Desktop 对话中直接调用本地工具,实现 AI 与系统的深度联动
  • 通过自然语言驱动复杂的多步骤自动化任务,代替繁琐手动操作
  • 将多个 MCP 工具组合使用,构建个人专属 AI 工作站

⚖️ 优点与不足

✅ 优点
  • +MPL-2.0 协议,可免费商用
  • +标准化 MCP 协议,生态互联性强
  • +与 Claude 官方生态无缝对接
  • +即插即用,配置简单快捷
⚠️ 不足
  • 依赖 Claude 客户端,非 Claude 用户无法使用
  • MCP 协议仍在持续演进,接口可能变更
  • 需要一定的配置步骤
⚠️ 使用须知

AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。

建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。

📄 License 说明

✅ MPL 2.0 — 文件级 Copyleft,修改的文件需开源,但可与闭源代码结合使用。

🔗 相关工具推荐

🧩 你可能还需要
基于当前 Skill 的能力图谱,自动补全的工具组合

❓ 常见问题 FAQ

UTCP是MCP的Go语言实现
💡 AI Skill Hub 点评

总体来看,UTCP工具 是一款质量优秀的MCP工具,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。

⬇️ 获取与下载
⬇ 下载源码 ZIP

✅ MPL-2.0 协议 · 可免费商用 · 直接从 aiskill88 服务器下载,无需跳转 GitHub

📚 深入学习 UTCP工具
查看分步骤安装教程和完整使用指南,快速上手这款工具
🌐 原始信息
原始名称 go-utcp
原始描述 开源MCP工具: Official Go implementation of the UTCP 。⭐119 · Go
Topics mcpaigolang
GitHub https://github.com/universal-tool-calling-protocol/go-utcp
License MPL-2.0
语言 Go
🔗 原始来源
🐙 GitHub 仓库  https://github.com/universal-tool-calling-protocol/go-utcp 🌐 官方网站  https://www.utcp.io/

收录时间:2026-07-13 · 更新时间:2026-07-13 · License:MPL-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。

📺 订阅 AI Skill Hub Daily Telegram 频道
每天 8 条精选 AI Skill、MCP、Agent 与自动化工具推送
加入频道 →