Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
제작자가 쓴 설명(영어 원문)
.NET assembly introspection MCP server with advanced reflection and type analysis capabilities
이 세 가지는 절대 하나의 점수로 합치지 않아요. 의견, 측정, 인기는 서로 다른 것이니까요.
이 서버는 아직 쉬운 말로 된 요약을 쓰지 못했어요. 아래는 레지스트리에서 읽어 온 사실이에요.
쓰고 있는 AI 앱을 고른 다음 설정을 복사하세요.
SHERLOCK_TOOL_PROFILE, SHERLOCK_SOURCE_FETCH어떤 앱을 쓰세요?
위에서 앱을 고르세요. 기억해 두었다가 다음부터는 버튼에 “…에 설치”라고 표시해요.
Claude Desktop → Settings → Developer → Edit Config를 열고 claude_desktop_config.json에 이 내용을 붙여 넣으세요
{
"mcpServers": {
"dotnet-sherlock": {
"command": "dnx",
"args": [
"Sherlock.MCP.Server",
"--yes"
],
"env": {
"SHERLOCK_TOOL_PROFILE": "your-value",
"SHERLOCK_SOURCE_FETCH": "your-value"
}
}
}
}터미널에서 이 명령어를 실행하세요
claude mcp add dotnet-sherlock --env SHERLOCK_TOOL_PROFILE=your-value --env SHERLOCK_SOURCE_FETCH=your-value -- dnx Sherlock.MCP.Server --yes실행하기 전에 빈 값을 채워 주세요.
프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요
{
"mcpServers": {
"dotnet-sherlock": {
"command": "dnx",
"args": [
"Sherlock.MCP.Server",
"--yes"
],
"env": {
"SHERLOCK_TOOL_PROFILE": "your-value",
"SHERLOCK_SOURCE_FETCH": "your-value"
}
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
프로젝트의 .vscode/mcp.json에 붙여 넣으세요
{
"servers": {
"dotnet-sherlock": {
"type": "stdio",
"command": "dnx",
"args": [
"Sherlock.MCP.Server",
"--yes"
],
"env": {
"SHERLOCK_TOOL_PROFILE": "your-value",
"SHERLOCK_SOURCE_FETCH": "your-value"
}
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
이 서버는 내 컴퓨터에서 실행되는데, ChatGPT는 인터넷에 있는 서버에만 연결할 수 있어요. 다른 앱을 고르세요.
선택한 앱은 이 기기의 쿠키에만 저장돼요. 계정은 필요 없어요.
SHERLOCK_TOOL_PROFILE envSHERLOCK_SOURCE_FETCH env이 설정은 제작자의 레지스트리 정보로 만들었어요. README와 다르면 README를 따르세요.
내 컴퓨터에서 실행돼요. 설치하면 AI 앱에 도구 목록이 나타나요.
누구나 볼 수 있는 정보를 점검했고, 코드는 읽지 않았어요. 해당하지 않는 항목은 개수에서 뺐어요. · 2026년 11월 7일 점검
레지스트리 이름이 GitHub 계정 jcucci 기준으로 확인되었고, 이 계정이 저장소를 소유하고 있어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
아직 GitHub에서 이 저장소를 읽어 오지 않았어요
내 컴퓨터에서 실행되는 서버라 인터넷 엔드포인트가 없어요
내 컴퓨터에서 실행되며, 대신 로컬 설정을 사용해요
내 컴퓨터에서 실행돼요. 도구 목록은 인터넷 엔드포인트가 있는 서버에서만 읽을 수 있어요
레지스트리에서 이 카테고리의 활성 서버 전체와 비교했어요. 스타는 게시자가 저장소를 소유한 경우에만 세요.
해당하는 기본 점검 1개 중 1개를 통과했어요. 코드는 읽지 않았어요. 키나 토큰을 넣기 전에 서버가 어떤 권한을 요구하는지 확인하고, 꼭 필요한 권한만 있는 키를 쓰세요.
레지스트리에 필수 키가 나와 있지 않아요. 다만 처음 쓸 때 로그인을 요구하는 서버도 있어요.
내 컴퓨터에서 실행돼요. AI 앱이 필요할 때 실행하고, 데이터는 제작자의 서버를 거치지 않아요(서버 자체가 온라인 서비스를 호출하는 경우는 제외).
GitHub 저장소는 마지막으로 – 변경됐고, 최신 버전(2.15.0)은 33일 전 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 .NET도 있어야 해요.
같은 카테고리에는: Spec Driven Development, Remote Desktop Commander, ProAgentStore
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
MCP-first control plane for ProAgentStore agents and private instances.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
Sherlock MCP for .NET is a comprehensive Model Context Protocol (MCP) server that provides deep introspection capabilities for .NET assemblies. It enables Language Learning Models (LLMs) to analyze and understand your .NET code with precision, delivering accurate and context-aware responses for complex development scenarios.
This tool is essential for developers who want to harness LLM capabilities for:
core profile whose other tools load on demand in groupsexplore_package, explain_type and who_calls prompts give one-click entry points into common analysis workflowsModelContextProtocol 2.1.0 (GA)2026-07-28, and negotiates down automatically for clients on earlier revisionsget_member_source returns a member's original source through its PDB or Source Link, decompile_member / decompile_type decompile to C#, and compare_api_surface diffs two assembly or package versions and flags breaking changes.find_endpoints, find_service_registrations, find_ef_entities and find_handlers list ASP.NET Core endpoints, DI registrations, EF Core entities and MediatR handlers.project.assets.json (get_package_graph shows the restored graph), and project tools evaluate projects with MSBuild.open_assembly handles, workflow prompts (explore_package, explain_type, who_calls), Mermaid diagrams for hierarchies and call graphs, and on-demand tool groups with discovery hints under the core profile.CHANGELOG.md for full details.With the .NET 10 SDK, dnx downloads the package from NuGet and runs it directly — no install step:
dnx -v q --yes Sherlock.MCP.Server@2.15.0
--yes skips the interactive confirmation prompt, which an MCP client launching the server over stdio can't answer. -v q stops dnx from printing a "Skipping NuGet package signature verification." notice to stdout the first time it downloads a version, which would otherwise corrupt the MCP stream and fail that first connection. Both are dnx options, so they must come before the package id; anything after it is passed to the server. Pinning the version keeps launches reproducible; bump it when you want to upgrade. The package is published with the McpServer package type, so it is also listed as an MCP server on NuGet.org.
On any supported SDK (.NET 8 or later), install the global tool from NuGet (adds sherlock-mcp to your PATH):
dotnet tool install -g Sherlock.MCP.Server
Alternatively, during development you can run the server locally:
dotnet run --project src/server/Sherlock.MCP.Server.csproj
Sherlock runs as a standard MCP server that communicates over stdio.
This repository is also a Claude Code plugin marketplace. The sherlock plugin bundles the server (launched with dnx, so it needs the .NET 10 SDK) and a skill that teaches the agent the locate → orient → drill-in → relationships workflow:
/plugin marketplace add jcucci/dotnet-sherlock-mcp
/plugin install sherlock@dotnet-sherlock-mcp
The plugin starts the server with the core tool profile, which matches the tools the skill covers. The agent can add the other tools by group with load_tools when it needs them (see Tool groups). To expose every tool from the start, set SHERLOCK_TOOL_PROFILE=full in the environment Claude Code is launched from.
Each user runs these commands once; restart Claude Code (or run /reload-plugins) afterwards. To pick up a new release, run /plugin marketplace update dotnet-sherlock-mcp. If you previously registered Sherlock with claude mcp add, remove that entry (claude mcp remove sherlock) so the tools aren't loaded twice.
To offer the plugin to everyone working in a repository, commit the following to that repository's .claude/settings.json. Claude Code prompts each teammate to install it when they trust the folder, so nobody has to type the commands above:
{
"extraKnownMarketplaces": { "dotnet-sherlock-mcp": { "source": { "source": "github", "repo": "jcucci/dotnet-sherlock-mcp" } } }, "enabledPlugins": { "sherlock@dotnet-sherlock-mcp": true } }
Teammates still need the .NET 10 SDK on their PATH for dnx.
claude mcp add sherlock -- dnx -v q --yes Sherlock.MCP.Server@2.15.0
.vscode/mcp.json):{
"servers": { "sherlock": { "type": "stdio", "command": "dnx", "args": ["-v", "q", "--yes", "Sherlock.MCP.Server@2.15.0"] } } }
sherlock-mcpsherlock-mcp command. Example JSON entry (refer to your client’s docs for exact file location/format):{
"servers": { "sherlock": { "command": "sherlock-mcp" } } }
No arguments are required. The server self-registers all tools when launched.
Large tool lists cost agents context and discoverability (Claude Code switches to deferred Tool Search once tool descriptions grow past roughly 10% of the context window). Sherlock can start with a smaller surface:
| Profile | Tools | Contents |
|---|---|---|
full (default) | 47 | Every tool, including the deprecated per-kind member tools |
core | 24 | load_tools and invoke_tool (see Tool groups), discovery (find_assembly_by_class_name, find_assembly_by_file_name, find_assembly_by_nuget_package, get_project_output_paths, open_assembly), orientation (get_assembly_info, get_types_from_assembly, get_type_info, get_type_hierarchy), members and docs (get_type_members, search_members, analyze_method, get_xml_docs_for_type, get_xml_docs_for_member) and relationships (find_implementations_of, find_methods_returning, find_extension_methods_for, find_references_to, get_method_calls), source (get_member_source, decompile_member) and API diffs (compare_api_surface) |
Select a profile with the --profile argument or the SHERLOCK_TOOL_PROFILE environment variable (the argument wins). An unknown profile name stops the server with an error.
{
"servers": { "sherlock": { "command": "sherlock-mcp", "args": ["--profile", "core"] // or: "env": { "SHERLOCK_TOOL_PROFILE": "core" } } } }
Under the core profile every other tool belongs to a group that the agent can load when it needs it:
| Group | Tools |
|---|---|
frameworks | find_endpoints, find_service_registrations, find_ef_entities, find_handlers |
project | analyze_solution, analyze_project, resolve_package_references, get_package_graph, find_deps_json_dependencies |
metadata | analyze_assembly, get_generic_type_info, get_nested_types, get_type_attributes, get_member_attributes, get_parameter_attributes |
decompile | decompile_type |
config | get_runtime_options, update_runtime_options |
legacy | the deprecated analyze_type, get_all_type_members and get_type_{methods,properties,fields,events,constructors} |
groups=['frameworks', …], it adds those tools to the server's tool list and returns each loaded tool's name, description and input schema. The server then sends notifications/tools/list_changed. Under protocol revision 2026-07-28 that only reaches clients that asked for it through subscriptions/listen.name with its arguments object and returns that tool's own result. Clients that never refresh their tool list can still use loaded groups this way. Calling a tool whose group isn't loaded fails with ToolNotLoaded and names the group.hints[], each { tool, reason, group }, pointing at a related tool. Examples:get_type_info on a type with nested types or generic parameters, or one that is a DbContext, controller or MediatR handler;find_implementations_of / find_references_to / find_extension_methods_for on those framework types or IServiceCollection;decompile_member → decompile_type;get_project_output_paths → analyze_project and get_package_graph.Dependency errors suggest get_package_graph. group is null for core tools.
core, tools/list is marked cacheScope: private, since a session's tool list grows as groups load. Under full, it stays public.The full profile lists every tool and omits load_tools and invoke_tool.
You usually don't need to paste anything. Sherlock ships its usage guidance in the MCP instructions field returned at initialize, and most MCP clients (including Claude Code) surface that to the agent automatically — so the guidance stays correct and versioned with the package, with no copy-paste to maintain.
The snippets below are optional reinforcement. Keep them short and principle-based rather than enumerating tool names and workflows: a static list pasted into your repo will drift as Sherlock's tools evolve, whereas the tools' own descriptions (and the server instructions) always match the version you're running.
Tool names are exposed in
snake_case(get_type_members,search_members, …); argument names stay camelCase (projection,nameContains).
If you installed the Claude Code plugin, its skill already covers this. Otherwise, you can add a short, optional pointer to your project's CLAUDE.md:
## .NET Assembly Analysis
Use the Sherlock MCP tools (get_type_members, search_members, …) for .NET type/assembly questions instead of guessing. Locate DLLs with the find_assembly_by_* / `get_project_outp
앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요