Context7
Up-to-date docs for AI agents
제작자가 쓴 설명(영어 원문)
Inspect and drive Flutter debug apps/games: semantic snapshots, search, and custom client tools
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.
필요한 것
어떤 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": {
"flutter-toolkit": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/arenukvern/flutter-mcp-toolkit:7.0.1"
]
}
}
}터미널에서 이 명령어를 실행하세요
claude mcp add flutter-toolkit -- docker run -i --rm ghcr.io/arenukvern/flutter-mcp-toolkit:7.0.1프로젝트의 .cursor/mcp.json 또는 ~/.cursor/mcp.json에 붙여 넣으세요
{
"mcpServers": {
"flutter-toolkit": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/arenukvern/flutter-mcp-toolkit:7.0.1"
]
}
}
}브라우저가 앱을 열지 물어본 다음, 앱이 설정을 보여 주고 확인을 요청해요. 키 값은 비워 두니 직접 입력하세요.
프로젝트의 .vscode/mcp.json에 붙여 넣으세요
{
"servers": {
"flutter-toolkit": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/arenukvern/flutter-mcp-toolkit:7.0.1"
]
}
}
}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.
이 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 Arenukvern 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(7.0.1)은 21일 전 레지스트리에 올라왔어요.
Claude Desktop, Claude Code, Cursor, VS Code처럼 MCP를 지원하는 앱에서 쓸 수 있어요. 내 컴퓨터에 Docker도 있어야 해요.
같은 카테고리에는: Context7, Codebase Memory, Exa
같은 카테고리에서 다른 제작자가 만든 서버예요. 스타 순으로 정렬했어요.
카드 보는 법:★ GitHub 스타i세 기둥: 리뷰 · 점검 · 사용 정도i업데이트 = 최신 버전을 레지스트리에 공개한 날i
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
제작자의 저장소에서 쓰인 그대로 가져왔어요. PopNx MCP가 검토하거나 보증한 내용이 아니에요. HTML, 이미지, 스크립트는 제거했어요. 링크를 누르면 이 사이트를 벗어나요.
Tool names on MCP are prefixed with `fmt_`, binaries and mcpServers keys were renamed — see [Migrating from v2.x to v3.0](../docs/start_here/migration_v2_to_v3.mdx).
This project now uses a shared core execution layer:
flutter-mcp-toolkit is the canonical command surface (connect, inspect, execute, diagnostics). fmtk is the short alias for the same CLI entrypoint.flutter-mcp-toolkit-server is a thin MCP protocol adapter that maps MCP tool/resource calls to the same core executor.tools/call namespace. Resource URIs are unchanged.The shared core module is available as flutter_mcp_core inside this package.
Use this sequence first on macOS:
mcp_toolkit to the app and call MCPToolkitBinding.instance.bootstrapFlutter(...).fmtk validate-runtime (or the canonical long form flutter-mcp-toolkit validate-runtime).fmt_list_client_tools_and_resources, fmt_client_tool, fmt_client_resource.Treat exec as expert mode. validate-runtime is the default first-pass proof command.
The CLI is now agent-first and exposes a canonical interface:
exec --name --args schema [--name ]capabilitiesserve (JSON-RPC 2.0 over stdio)snapshot create --name [--args ]snapshot diff --from --to bundle create --from-snapshot [--output ]doctor [--json] [--target ] [--timeout-ms ]permissions status|request|open-settings [--kind visual_capture]Safe-write flags for write-producing commands:
snapshot create: --check --diff --backup --no-overwritebundle create: --check --diff --backup --no-overwriteexec targets commands in the shared CommandCatalog (CLI command names may be bare because they appear only after --name; the CLI is not the MCP tools/call namespace): connect, session_start, session_exec, session_end, diagnose, watch, explain_errors, status, discover_debug_apps, get_vm, get_extension_rpcs, hot_reload_flutter, hot_restart_flutter, get_active_ports, get_app_errors, get_screenshots, focus_window, get_view_details, inspect_widget_at_point, capture_ui_snapshot, debug_dump_layer_tree, debug_dump_semantics_tree, debug_dump_render_tree, debug_dump_focus_tree, fmt_list_client_tools_and_resources, fmt_client_tool, fmt_client_resource, dynamicRegistryStats, semantic_snapshot, tap_widget, long_press, enter_text, reveal_search, scroll, swipe, drag, hot_reload_and_capture, evaluate_dart_expression, get_recent_logs.
MCP names. When invoked via MCP
tools/call, exposed tools use the
fmt_capability prefix (e.g.fmt_tap_widget,
fmt_hot_reload_and_capture). The dynamic-registry host trio is exposed as
fmt_list_client_tools_and_resources,fmt_client_tool, and
fmt_client_resourcewhere enabled.dynamicRegistryStatsremains available
from the CLI catalog for low-level diagnostics and is not part of the default
MCP tool surface.
Interaction tools (catalog names: semantic_snapshot → tap_widget / enter_text / reveal_search / scroll / swipe / long_press / drag) follow a Playwright-style ref model: take a snapshot, then pass ref: "s_N" (and optional snapshotId for staleness detection) into the interaction tool. reveal_search is the bounded helper for off-screen semantic targets: it snapshots, matches one narrow selector, scrolls up to maxAttempts, and returns a fresh ref/snapshotId plus trace. hot_reload_and_capture fuses reload + screenshot + fresh snapshot + errors. evaluate_dart_expression runs an ad-hoc Dart expression against the app's root library. See docs/start_here/cli_quick_recipes.mdx and docs/guides/interaction_cookbook.mdx for the full surface and golden paths.
Packaged installs include both flutter-mcp-toolkit and fmtk; examples below use dart run for source-tree development. Prefer fmtk for compact local loops and keep flutter-mcp-toolkit in onboarding, install, PATH, and MCP config docs where the searchable canonical name helps.
CLI runs the same shared command catalog/executor as MCP. Preferred debugging path: discover_debug_apps -> capture_ui_snapshot -> inspect_widget_at_point. get_active_ports and dynamicRegistryStats remain available in CLI for low-level diagnostics, but are intentionally not MCP-exposed by default.
# introspection
dart run bin/flutter_mcp_toolkit.dart schema dart run bin/flutter_mcp_toolkit.dart capabilities
dart run bin/flutter_mcp_toolkit.dart exec --name status --args '{}' dart run bin/flutter_mcp_toolkit.dart exec --name get_vm --args '{}' dart run bin/flutter_mcp_toolkit.dart exec --name get_vm --args '{"connection":{"targetId":"ws://127.0.0.1:8181/ /ws"}}' dart run bin/flutter_mcp_toolkit.dart exec --name discover_debug_apps --args '{}' dart run bin/flutter_mcp_toolkit.dart exec --name capture_ui_snapshot --args '{"connection":{"targetId":"ws://127.0.0.1:8181/ /ws"}}' dart run bin/flutter_mcp_toolkit.dart exec --name inspect_widget_at_point --args '{"x":120,"y":220,"connection":{"targetId":"ws://127.0.0.1:8181/ /ws"}}'
dart run bin/flutter_mcp_toolkit.dart exec --name session_start --args '{"mode":"uri","uri":"ws://127.0.0.1:8181/ /ws"}' dart run bin/flutter_mcp_toolkit.dart exec --name session_exec --args '{"command":"get_app_errors","arguments":{"count":4}}' dart run bin/flutter_mcp_toolkit.dart exec --name session_end --args '{}'
dart run bin/flutter_mcp_toolkit.dart snapshot create --name baseline --args '{"commands":[{"name":"status","args":{}}]}' --check --diff dart run bin/flutter_mcp_toolkit.dart snapshot diff --from baseline --to after_fix dart run bin/flutter_mcp_toolkit.dart bundle create --from-snapshot baseline --backup
dart run bin/flutter_mcp_toolkit.dart doctor --json dart run bin/flutter_mcp_toolkit.dart permissions status dart run bin/flutter_mcp_toolkit.dart permissions request dart run bin/flutter_mcp_toolkit.dart permissions open-settings dart run bin/flutter_mcp_toolkit.dart exec --name get_extension_rpcs --args '{}'
dart run bin/flutter_mcp_toolkit.dart exec --name get_screenshots --args '{"mode":"desktop_window","permissionPolicy":"auto_request_once"}' dart run bin/flutter_mcp_toolkit.dart exec --name capture_ui_snapshot --args '{"screenshotMode":"auto","permissionPolicy":"auto_request_once"}'
dart run bin/flutter_mcp_toolkit.dart --save-images --output-dir .flutter_mcp/app validate-runtime \ --target ws://127.0.0.1:8181/ /ws \ --timeout-ms 10000 \ --post-reload-delay-ms 500 \ --after-reload
dart run bin/flutter_mcp_toolkit.dart --vm-service-uri ws://127.0.0.1:8181/ /ws validate-runtime \ --timeout-ms 10000
dart run bin/flutter_mcp_toolkit.dart validate-runtime \ --target ws://127.0.0.1:8181/ /ws \ --install-skill
CLI runtime gate for app inspection:
ext.mcp.toolkit.app_errors, ext.mcp.toolkit.view_details, ext.mcp.toolkit.view_screenshots, and ext.mcp.toolkit.inspect_widget_at_point from get_extension_rpcs.mcp_toolkit is installed, initialized, and the app is hot restarted or rerun.captureHints on get_view_details or screenshot payloads (platformViewsDetected or weakSignalsDetected). Use mode: desktop_window (or keep auto, which upgrades on macOS host when strong signals are present) instead of flutter_layer.desktopCaptureRetried in the payload). You can also call focus_window then retry get_screenshots with desktop_window.doctor --json --target --timeout-ms 10000.--output-dir is set, validate-runtime mirrors its JSON envelope to /validate-runtime.json and screenshot files are written under /.mcp_screenshots/.validate-runtime tries host desktop_window first when auto selects it. Executor recovery retries focus+capture once (desktopCaptureRetried). If desktop_window still fails, validate-runtime retries once with flutter_layer (including when platform views are detected, e.g. the showcase AppKitView). Check data.summary.capturePlatformViewsDetected, captureFocusAttempted, and captureFallbackUsed. Web headful capture options: ADR 0007.doctor stays read-only. It now reports visual_capture_backend, visual_capture_permission, visual_capture_truth_mode, and app_permission_bridge.auto_request_once for exec get_screenshots, exec capture_ui_snapshot, and validate-runtime. Raw command schemas still default to check_only.desktop_window (including iOS Simulator and Chrome / web on a Mac host). Screen Recording permission belongs to the host process running flutter-mcp-toolkit, not the Flutter app. Use permissions request for the native prompt and permissions open-settings after a denial.get_view_details and view_screenshots include captureHints when native platform views are detected (AndroidView, UiKitView, AppKitView, HtmlElementView, PlatformViewLink). Texture sets weakSignalsDetected with a soft warning (no auto upgrade). Hybrid engines without platform views can use MCPToolkitBinding.captureHintsContributor in the app.get_screenshots MCP responses include routing metadata in meta and a leading JSON text block when captureHints / warnings are present.AppKitView on macOS (make showcase), HtmlElementView on web (flutter run -d chrome). Use make showcase-stop before a new macOS run.chrome / web / web-server): desktop_window uses macOS ScreenCaptureKit → Chrome CDP (Page.captureScreenshot) → flutter_layer. Global flags: --web-browser-debugging-port, --web-port. Opt-in live test: RUN_WEB_CDP_INTEGRATION=1 dart test test/web_cdp_integration_test.dart. See ADR 0007.permissions or doctor can verify bridge-backed permission tools/resources. Use --target or the global --vm-service-uri when probing those platforms.desktop_window never silently falls back. auto may fall back to flutter_layer, but responses always report requestedMode, actualMode, permissionStatus, and fallbackReason.Troubleshooting:
flutter-mcp-toolkit permissions status first. If status is still denied, open System Settings from the CLI and grant Screen Recording to the terminal or client process you are using.doctor --json shows visual_capture_truth_mode=flutter_layer on macOS, you are not getting native window p앞부분만 보여 드려요. 나머지는 저장소에서 읽어 보세요