DotnetTemplateMCP
MCP server wrapping the .NET Template Engine for AI-driven template discovery, inspection, and instantiation
README
Microsoft.TemplateEngine.MCP
<!-- mcp-name: io.github.YuliiaKovalova/dotnet-template-mcp -->
An MCP server that lets AI agents work with dotnet new templates — search, inspect, preview, and create projects through natural conversation instead of memorizing CLI flags.
<a href="https://glama.ai/mcp/servers/@YuliiaKovalova/dotnet-template-mcp"> <img width="380" height="200" src="https://glama.ai/mcp/servers/@YuliiaKovalova/dotnet-template-mcp/badge" alt="DotnetTemplateMCP MCP server" /> </a>
Instead of this:
dotnet new list --language C#
dotnet new webapi --help
dotnet new webapi --auth Individual --use-controllers --name MyApi --output ./MyApi
Your AI agent just says: "I need a web API with authentication and controllers" — and the MCP server figures out the rest.
Template Validation for Authors
Building a custom dotnet new template? template_validate catches mistakes before you publish — no more guessing if your template.json is correct:
Agent calls: template_validate("./my-template")
← Returns:
{
"valid": false,
"summary": "2 error(s), 1 warning(s), 3 suggestion(s)",
"errors": [
"Missing required field 'shortName'.",
"Parameter 'Framework': default value 'net7.0' is not in the choices list."
],
"warnings": [
"Missing 'sourceName'. Without it, the generated project name won't be customizable via --name."
],
"suggestions": [
"Consider adding a 'description' field to help users understand what this template creates.",
"Consider adding 'language' tag (e.g., 'C#') for better discoverability.",
"Consider adding 'type' tag (e.g., 'project', 'item') for filtering."
]
}
What it catches: missing required fields, invalid identity format, short name conflicts with CLI commands, parameter issues (missing defaults, empty choices, prefix collisions, type mismatches), broken computed symbols, constraint misconfiguration, and missing tags.
No existing tooling does this — most template authors discover issues only after dotnet new install fails or produces wrong output.
Tools
| Tool | What it does |
|---|---|
template_search |
Search locally and on NuGet.org — one call, ranked results |
template_list |
List what's installed, filter by language/type/classification |
template_inspect |
Parameters, constraints, post-actions — all in one shot |
template_instantiate |
Create a project. Not installed? Auto-resolves from NuGet. Elicits missing params interactively |
template_dry_run |
Preview files without touching disk |
template_install |
Install a package (idempotent — skips if already there) |
template_uninstall |
Remove a template package |
templates_installed |
Inventory of everything installed |
template_from_intent |
"web API with auth" → webapi + auth=Individual — no LLM needed |
template_create_from_existing |
Analyze a .csproj → generate a reusable template matching repo conventions |
template_compose |
Execute a sequence of templates (project + items) in one workflow |
template_suggest_parameters |
Suggest parameter values with rationale based on cross-parameter relationships |
template_validate |
Validate a local template directory for authoring issues before publishing |
solution_analyze |
Analyze a solution/workspace — project structure, frameworks, CPM status |
Quick Start
Global tool (.NET 8+)
dotnet tool install --global DotnetTemplateMCP --version 1.2.0
Zero-install with dnx (.NET 10+)
dnx -y DotnetTemplateMCP --version 1.2.0
VS Code / GitHub Copilot
Add to mcp.json:
{
"servers": {
"dotnet-templates": {
"type": "stdio",
"command": "dnx",
"args": ["-y", "DotnetTemplateMCP", "--version", "1.2.0"]
}
}
}
📖 Claude Desktop, Cursor, and more →
Transport Modes
Stdio (default)
Standard I/O transport for local CLI and tool usage:
template-engine-mcp # stdio is the default
template-engine-mcp --transport stdio # explicit
HTTP (remote / cloud / team-shared)
Streamable HTTP transport for remote, multi-tenant, or CI/CD deployment:
template-engine-mcp --transport http
# or via environment variable:
MCP_TEMPLATE_TRANSPORT=http template-engine-mcp
The HTTP server exposes:
/mcp— MCP streamable HTTP endpoint/health— Health check endpoint
Configure the listen URL:
MCP_TEMPLATE_HTTP_URL=http://0.0.0.0:8080 template-engine-mcp --transport http
Connect your MCP client:
{
"servers": {
"dotnet-templates": {
"type": "http",
"url": "http://localhost:5005/mcp"
}
}
}
Interactive Elicitation
When a template has required parameters that weren't provided, the server asks the user interactively via MCP elicitation — instead of failing. Template parameter types are mapped to form fields:
| Template Parameter | Elicitation Field |
|---|---|
string |
Text input |
bool / boolean |
Checkbox |
int / number |
Number input |
| Choice parameter | Single-select dropdown |
Disable with MCP_TEMPLATE_ELICITATION=false.
How it works
You: "I need a web API with authentication, controllers, and Docker support"
→ template_from_intent extracts keywords: web api, authentication, controllers, docker
→ Matches: webapi (confidence: 0.85)
→ Resolves: auth=Individual, UseControllers=true, EnableDocker=true
→ template_instantiate creates the project
The server also does smart defaults (AOT → latest framework, auth → HTTPS stays on), parameter validation before writing files, constraint checking (OS, SDK, workload), interactive elicitation of missing required parameters, and auto-resolves templates from NuGet if they're not installed.
CPM & Latest Package Versions
When creating a project inside a solution that uses Central Package Management, the server automatically:
- Detects
Directory.Packages.propsby walking up the directory tree - Strips
Versionattributes from generated.csprojPackageReferences - Adds missing
<PackageVersion>entries toDirectory.Packages.props - Resolves latest stable NuGet versions — no more stale hardcoded versions from templates
Before (what dotnet new generates):
<PackageReference Include="Serilog" Version="3.1.0" /> ← stale, breaks CPM
After (what template_instantiate produces):
.csproj: <PackageReference Include="Serilog" />
Directory.Packages.props: <PackageVersion Include="Serilog" Version="4.2.0" />
Works for standalone projects too — versions are updated directly in the .csproj.
Multi-Template Composition
Chain multiple templates in one call with template_compose:
[
{"templateName": "webapi", "name": "MyApi", "parametersJson": "{\"auth\": \"Individual\"}"},
{"templateName": "gitignore", "target": "."}
]
📖 Architecture & smart behaviors →
Tool Profiles (Lite vs Full)
By default, all 13 tools are available. If your agent works better with fewer tools, set the MCP_TEMPLATE_TOOL_PROFILE environment variable:
| Profile | Tools | When to use |
|---|---|---|
full (default) |
All 13 tools | Full control — advanced workflows, composition, custom templates |
lite |
5 core tools | Simpler agents that just need to find and create projects |
Lite profile tools: template_from_intent, template_instantiate, template_inspect, template_search, template_dry_run
{
"servers": {
"dotnet-template-mcp": {
"command": "dotnet-template-mcp",
"env": {
"MCP_TEMPLATE_TOOL_PROFILE": "lite"
}
}
}
}
Non-lite tools will return a helpful message explaining they're disabled and how to enable them.
Documentation
| Doc | What's in it |
|---|---|
| Configuration | VS Code, Claude Desktop, Cursor setup + troubleshooting |
| Tool Reference | Every tool's parameters, types, and examples |
| Architecture | Template cache, smart behaviors, telemetry, project structure |
| MCP vs Skills | Why MCP over Copilot Skills — benefits and downsides |
| Plain LLM vs MCP | Side-by-side: what a plain LLM does vs. the MCP tool (4 scenarios) |
| Skills Equivalent | What it'd take to cover this with Skills instead |
Building & Testing
dotnet build
dotnet test # 185+ tests — unit, integration, and E2E
CI runs on push/PR via GitHub Actions (Ubuntu + Windows).
Contributing
Contributions are welcome! Please open an issue to discuss proposed changes before submitting a PR.
# Setup
dotnet restore
dotnet build
# Run tests
dotnet test
# Pack locally
dotnet pack src/Microsoft.TemplateEngine.MCP -o nupkg/
Changelog
See CHANGELOG.md for release history.
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。