@midscene/shared 1.9.8 → 1.10.1-beta-20260624112700.0

package/dist/types/{mcp → agent-tools}/tool-defaults.d.ts

@@ -1,10 +1,10 @@
 /**
  * Unified, declarative mechanism for "force a default option on every tool
- * call" behaviors exposed by MCP servers and the device / Agent Skill CLIs.
+ * call" behaviors exposed by device and Agent Skill CLIs.
  *
  * Adding a new behavior flag (e.g. `--deep-search`) is a one-line change to
  * {@link TOOL_BEHAVIOR_FLAGS}: declare which default-option "bag" it fills.
- * The tool generator, servers, tools managers and CLI parsing are all generic
+ * The tool generator, tools managers and CLI parsing are all generic
  * over {@link ToolDefaults} and never need to learn about individual flags.
  *
  * See https://github.com/web-infra-dev/midscene/issues/2446.
@@ -50,13 +50,12 @@ export declare function resolveToolDefaults(isEnabled: (cli: string) => boolean)
  *
  * Behavior flags (e.g. `--deep-locate`) are global: they may appear anywhere
  * in argv and are not tied to a specific sub-command. They are recognized by
- * exact kebab-case match — the same surface the MCP `parseArgs` config exposes
- * — and removed so a strict per-command parser never sees them. Every other
+ * exact kebab-case match and removed so a strict per-command parser never sees them. Every other
  * token is returned untouched and in order for that per-command parser.
  *
  * This is the single place that knows how a behavior flag looks on the command
- * line; both the device / Agent Skill CLI and the MCP launch path resolve their
- * defaults from {@link TOOL_BEHAVIOR_FLAGS} through here / {@link resolveToolDefaults}.
+ * line; the device / Agent Skill CLI resolves defaults from
+ * {@link TOOL_BEHAVIOR_FLAGS} through here / {@link resolveToolDefaults}.
  */
 export declare function stripBehaviorFlags(argv: readonly string[]): {
     rawArgs: string[];

package/dist/types/{mcp → agent-tools}/tool-generator.d.ts

@@ -3,7 +3,7 @@ import type { ActionSpaceItem, BaseAgent, ToolCliMetadata, ToolDefinition, ToolS
 import { composeUserPrompt } from './user-prompt';
 export { composeUserPrompt };
 /**
- * Converts DeviceAction from actionSpace into MCP ToolDefinition
+ * Converts DeviceAction from actionSpace into ToolDefinition.
  * This is the core logic that removes need for hardcoded tool definitions
  */
 export declare function generateToolsFromActionSpace(actionSpace: ActionSpaceItem[], getAgent: (args?: Record<string, unknown>) => Promise<BaseAgent>, sanitizeArgs?: (args: Record<string, unknown>) => Record<string, unknown>, initArgSchema?: ToolSchema, initArgCliMetadata?: ToolCliMetadata, toolDefaults?: ToolDefaults): ToolDefinition[];

package/dist/types/{mcp → agent-tools}/types.d.ts

@@ -1,4 +1,3 @@
-import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
 import type { z } from 'zod';
 import type { ToolDefaults } from './tool-defaults';
 /**
@@ -7,7 +6,7 @@ import type { ToolDefaults } from './tool-defaults';
 export declare const defaultAppLoadingTimeoutMs = 10000;
 export declare const defaultAppLoadingCheckIntervalMs = 2000;
 /**
- * Content item types for tool results (MCP compatible)
+ * Content item types for tool results.
  */
 export type ToolResultContent = {
     type: 'text';
@@ -33,7 +32,7 @@ export type ToolResultContent = {
     };
 };
 /**
- * Result type for tool execution (MCP compatible)
+ * Result type for tool execution.
  */
 export interface ToolResult {
     [x: string]: unknown;
@@ -58,7 +57,7 @@ export interface ToolCliMetadata {
     options?: Record<string, ToolCliOption>;
 }
 /**
- * Tool definition for MCP server
+ * Tool definition for Midscene CLI and Skill surfaces.
  */
 export interface ToolDefinition<T = Record<string, unknown>> {
     name: string;
@@ -67,9 +66,6 @@ export interface ToolDefinition<T = Record<string, unknown>> {
     handler: ToolHandler<T>;
     cli?: ToolCliMetadata;
 }
-/**
- * Tool type for mcpKitForAgent return value
- */
 export type Tool = ToolDefinition;
 /**
  * Action space item definition
@@ -137,10 +133,9 @@ export interface BaseDevice {
     destroy?(): Promise<void>;
 }
 /**
- * Interface for platform-specific MCP tools manager
+ * Interface for platform-specific tools manager.
  */
 export interface IMidsceneTools {
-    attachToServer(server: McpServer): void;
     initTools(): Promise<void>;
     destroy?(): Promise<void>;
     setToolDefaults?(toolDefaults: ToolDefaults): void;

package/dist/types/cli/cli-args.d.ts

@@ -1,4 +1,4 @@
-import type { ToolCliOption, ToolDefinition } from '../mcp/types';
+import type { ToolCliOption, ToolDefinition } from '../agent-tools/types';
 export declare function parseValue(raw: string): unknown;
 export declare function parseCliArgs(args: string[]): Record<string, unknown>;
 export declare function getCliOptionDisplay(key: string, cliOption?: ToolCliOption): {

package/dist/types/cli/cli-runner.d.ts

@@ -1,5 +1,5 @@
-import type { BaseMidsceneTools } from '../mcp/base-tools';
-import type { ToolDefinition } from '../mcp/types';
+import type { BaseMidsceneTools } from '../agent-tools/base-tools';
+import type { ToolDefinition } from '../agent-tools/types';
 export interface CLIExtraCommand {
     name: string;
     def: ToolDefinition;

package/dist/types/env/types.d.ts

@@ -5,9 +5,11 @@ export declare const MIDSCENE_DEBUG_MODEL_PROFILE = "MIDSCENE_DEBUG_MODEL_PROFIL
 export declare const MIDSCENE_DEBUG_MODEL_RESPONSE = "MIDSCENE_DEBUG_MODEL_RESPONSE";
 export declare const MIDSCENE_DANGEROUSLY_PRINT_ALL_CONFIG = "MIDSCENE_DANGEROUSLY_PRINT_ALL_CONFIG";
 export declare const MIDSCENE_DEBUG_MODE = "MIDSCENE_DEBUG_MODE";
-export declare const MIDSCENE_MCP_USE_PUPPETEER_MODE = "MIDSCENE_MCP_USE_PUPPETEER_MODE";
+export declare const MIDSCENE_CHROME_PATH = "MIDSCENE_CHROME_PATH";
+/**
+ * @deprecated Use MIDSCENE_CHROME_PATH instead. This is kept for backward compatibility.
+ */
 export declare const MIDSCENE_MCP_CHROME_PATH = "MIDSCENE_MCP_CHROME_PATH";
-export declare const MIDSCENE_MCP_ANDROID_MODE = "MIDSCENE_MCP_ANDROID_MODE";
 export declare const DOCKER_CONTAINER = "DOCKER_CONTAINER";
 export declare const MIDSCENE_LANGSMITH_DEBUG = "MIDSCENE_LANGSMITH_DEBUG";
 export declare const MIDSCENE_LANGFUSE_DEBUG = "MIDSCENE_LANGFUSE_DEBUG";
@@ -104,22 +106,22 @@ export declare const UNUSED_ENV_KEYS: string[];
  * can not be override by overrideAIConfig
  */
 export declare const BASIC_ENV_KEYS: readonly ["MIDSCENE_DEBUG_MODE", "MIDSCENE_DEBUG_MODEL_PROFILE", "MIDSCENE_DEBUG_MODEL_RESPONSE", "MIDSCENE_RUN_DIR"];
-export declare const BOOLEAN_ENV_KEYS: readonly ["MIDSCENE_CACHE", "MIDSCENE_MCP_USE_PUPPETEER_MODE", "MIDSCENE_MCP_ANDROID_MODE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET"];
+export declare const BOOLEAN_ENV_KEYS: readonly ["MIDSCENE_CACHE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET"];
 export declare const NUMBER_ENV_KEYS: readonly ["MIDSCENE_CACHE_MAX_FILENAME_LENGTH", "MIDSCENE_REPLANNING_CYCLE_LIMIT"];
-export declare const STRING_ENV_KEYS: readonly ["MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER"];
+export declare const STRING_ENV_KEYS: readonly ["MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_CHROME_PATH", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER"];
 /**
  * Non model related env keys, used for globally controlling the behavior of midscene
  * Can not be override by agent.modelConfig but can be override by overrideAIConfig
  * Can be access at any time
  */
-export declare const GLOBAL_ENV_KEYS: readonly ["MIDSCENE_CACHE", "MIDSCENE_MCP_USE_PUPPETEER_MODE", "MIDSCENE_MCP_ANDROID_MODE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET", "MIDSCENE_CACHE_MAX_FILENAME_LENGTH", "MIDSCENE_REPLANNING_CYCLE_LIMIT", "MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER"];
+export declare const GLOBAL_ENV_KEYS: readonly ["MIDSCENE_CACHE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET", "MIDSCENE_CACHE_MAX_FILENAME_LENGTH", "MIDSCENE_REPLANNING_CYCLE_LIMIT", "MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_CHROME_PATH", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER"];
 /**
  * Model related eve keys, used for declare which model to use.
  * Can be override by both agent.modelConfig and overrideAIConfig
  * Can only be access after agent.constructor
  */
 export declare const MODEL_ENV_KEYS: readonly ["MIDSCENE_MODEL_NAME", "MIDSCENE_MODEL_INIT_CONFIG_JSON", "MIDSCENE_MODEL_EXTRA_BODY_JSON", "MIDSCENE_MODEL_API_KEY", "MIDSCENE_MODEL_BASE_URL", "MIDSCENE_MODEL_SOCKS_PROXY", "MIDSCENE_MODEL_HTTP_PROXY", "MIDSCENE_MODEL_TIMEOUT", "MIDSCENE_MODEL_TEMPERATURE", "MIDSCENE_MODEL_RETRY_COUNT", "MIDSCENE_MODEL_RETRY_INTERVAL", "MIDSCENE_MODEL_REASONING_EFFORT", "MIDSCENE_MODEL_REASONING_ENABLED", "MIDSCENE_MODEL_REASONING_BUDGET", "MIDSCENE_USE_VLM_UI_TARS", "MIDSCENE_USE_QWEN_VL", "MIDSCENE_USE_QWEN3_VL", "MIDSCENE_USE_DOUBAO_VISION", "MIDSCENE_USE_GEMINI", "MIDSCENE_USE_VL_MODEL", "OPENAI_API_KEY", "OPENAI_BASE_URL", "MIDSCENE_OPENAI_INIT_CONFIG_JSON", "MIDSCENE_OPENAI_HTTP_PROXY", "MIDSCENE_OPENAI_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_NAME", "MIDSCENE_INSIGHT_MODEL_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_HTTP_PROXY", "MIDSCENE_INSIGHT_MODEL_BASE_URL", "MIDSCENE_INSIGHT_MODEL_API_KEY", "MIDSCENE_INSIGHT_MODEL_INIT_CONFIG_JSON", "MIDSCENE_INSIGHT_MODEL_EXTRA_BODY_JSON", "MIDSCENE_INSIGHT_MODEL_TIMEOUT", "MIDSCENE_INSIGHT_MODEL_TEMPERATURE", "MIDSCENE_INSIGHT_MODEL_RETRY_COUNT", "MIDSCENE_INSIGHT_MODEL_RETRY_INTERVAL", "MIDSCENE_INSIGHT_MODEL_FAMILY", "MIDSCENE_INSIGHT_MODEL_REASONING_EFFORT", "MIDSCENE_INSIGHT_MODEL_REASONING_ENABLED", "MIDSCENE_INSIGHT_MODEL_REASONING_BUDGET", "MIDSCENE_PLANNING_MODEL_NAME", "MIDSCENE_PLANNING_MODEL_SOCKS_PROXY", "MIDSCENE_PLANNING_MODEL_HTTP_PROXY", "MIDSCENE_PLANNING_MODEL_BASE_URL", "MIDSCENE_PLANNING_MODEL_API_KEY", "MIDSCENE_PLANNING_MODEL_INIT_CONFIG_JSON", "MIDSCENE_PLANNING_MODEL_EXTRA_BODY_JSON", "MIDSCENE_PLANNING_MODEL_TIMEOUT", "MIDSCENE_PLANNING_MODEL_TEMPERATURE", "MIDSCENE_PLANNING_MODEL_RETRY_COUNT", "MIDSCENE_PLANNING_MODEL_RETRY_INTERVAL", "MIDSCENE_PLANNING_MODEL_FAMILY", "MIDSCENE_PLANNING_MODEL_REASONING_EFFORT", "MIDSCENE_PLANNING_MODEL_REASONING_ENABLED", "MIDSCENE_PLANNING_MODEL_REASONING_BUDGET", "MIDSCENE_MODEL_FAMILY"];
-export declare const ALL_ENV_KEYS: readonly [...string[], "MIDSCENE_DEBUG_MODE", "MIDSCENE_DEBUG_MODEL_PROFILE", "MIDSCENE_DEBUG_MODEL_RESPONSE", "MIDSCENE_RUN_DIR", "MIDSCENE_CACHE", "MIDSCENE_MCP_USE_PUPPETEER_MODE", "MIDSCENE_MCP_ANDROID_MODE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET", "MIDSCENE_CACHE_MAX_FILENAME_LENGTH", "MIDSCENE_REPLANNING_CYCLE_LIMIT", "MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER", "MIDSCENE_MODEL_NAME", "MIDSCENE_MODEL_INIT_CONFIG_JSON", "MIDSCENE_MODEL_EXTRA_BODY_JSON", "MIDSCENE_MODEL_API_KEY", "MIDSCENE_MODEL_BASE_URL", "MIDSCENE_MODEL_SOCKS_PROXY", "MIDSCENE_MODEL_HTTP_PROXY", "MIDSCENE_MODEL_TIMEOUT", "MIDSCENE_MODEL_TEMPERATURE", "MIDSCENE_MODEL_RETRY_COUNT", "MIDSCENE_MODEL_RETRY_INTERVAL", "MIDSCENE_MODEL_REASONING_EFFORT", "MIDSCENE_MODEL_REASONING_ENABLED", "MIDSCENE_MODEL_REASONING_BUDGET", "MIDSCENE_USE_VLM_UI_TARS", "MIDSCENE_USE_QWEN_VL", "MIDSCENE_USE_QWEN3_VL", "MIDSCENE_USE_DOUBAO_VISION", "MIDSCENE_USE_GEMINI", "MIDSCENE_USE_VL_MODEL", "OPENAI_API_KEY", "OPENAI_BASE_URL", "MIDSCENE_OPENAI_INIT_CONFIG_JSON", "MIDSCENE_OPENAI_HTTP_PROXY", "MIDSCENE_OPENAI_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_NAME", "MIDSCENE_INSIGHT_MODEL_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_HTTP_PROXY", "MIDSCENE_INSIGHT_MODEL_BASE_URL", "MIDSCENE_INSIGHT_MODEL_API_KEY", "MIDSCENE_INSIGHT_MODEL_INIT_CONFIG_JSON", "MIDSCENE_INSIGHT_MODEL_EXTRA_BODY_JSON", "MIDSCENE_INSIGHT_MODEL_TIMEOUT", "MIDSCENE_INSIGHT_MODEL_TEMPERATURE", "MIDSCENE_INSIGHT_MODEL_RETRY_COUNT", "MIDSCENE_INSIGHT_MODEL_RETRY_INTERVAL", "MIDSCENE_INSIGHT_MODEL_FAMILY", "MIDSCENE_INSIGHT_MODEL_REASONING_EFFORT", "MIDSCENE_INSIGHT_MODEL_REASONING_ENABLED", "MIDSCENE_INSIGHT_MODEL_REASONING_BUDGET", "MIDSCENE_PLANNING_MODEL_NAME", "MIDSCENE_PLANNING_MODEL_SOCKS_PROXY", "MIDSCENE_PLANNING_MODEL_HTTP_PROXY", "MIDSCENE_PLANNING_MODEL_BASE_URL", "MIDSCENE_PLANNING_MODEL_API_KEY", "MIDSCENE_PLANNING_MODEL_INIT_CONFIG_JSON", "MIDSCENE_PLANNING_MODEL_EXTRA_BODY_JSON", "MIDSCENE_PLANNING_MODEL_TIMEOUT", "MIDSCENE_PLANNING_MODEL_TEMPERATURE", "MIDSCENE_PLANNING_MODEL_RETRY_COUNT", "MIDSCENE_PLANNING_MODEL_RETRY_INTERVAL", "MIDSCENE_PLANNING_MODEL_FAMILY", "MIDSCENE_PLANNING_MODEL_REASONING_EFFORT", "MIDSCENE_PLANNING_MODEL_REASONING_ENABLED", "MIDSCENE_PLANNING_MODEL_REASONING_BUDGET", "MIDSCENE_MODEL_FAMILY"];
+export declare const ALL_ENV_KEYS: readonly [...string[], "MIDSCENE_DEBUG_MODE", "MIDSCENE_DEBUG_MODEL_PROFILE", "MIDSCENE_DEBUG_MODEL_RESPONSE", "MIDSCENE_RUN_DIR", "MIDSCENE_CACHE", "MIDSCENE_LANGSMITH_DEBUG", "MIDSCENE_LANGFUSE_DEBUG", "MIDSCENE_REPORT_QUIET", "MIDSCENE_CACHE_MAX_FILENAME_LENGTH", "MIDSCENE_REPLANNING_CYCLE_LIMIT", "MIDSCENE_ADB_PATH", "MIDSCENE_ADB_REMOTE_HOST", "MIDSCENE_ADB_REMOTE_PORT", "MIDSCENE_ANDROID_IME_STRATEGY", "MIDSCENE_IOS_DEVICE_UDID", "MIDSCENE_IOS_SIMULATOR_UDID", "MIDSCENE_REPORT_TAG_NAME", "MIDSCENE_PREFERRED_LANGUAGE", "MATCH_BY_POSITION", "MIDSCENE_CHROME_PATH", "MIDSCENE_MCP_CHROME_PATH", "DOCKER_CONTAINER", "MIDSCENE_MODEL_NAME", "MIDSCENE_MODEL_INIT_CONFIG_JSON", "MIDSCENE_MODEL_EXTRA_BODY_JSON", "MIDSCENE_MODEL_API_KEY", "MIDSCENE_MODEL_BASE_URL", "MIDSCENE_MODEL_SOCKS_PROXY", "MIDSCENE_MODEL_HTTP_PROXY", "MIDSCENE_MODEL_TIMEOUT", "MIDSCENE_MODEL_TEMPERATURE", "MIDSCENE_MODEL_RETRY_COUNT", "MIDSCENE_MODEL_RETRY_INTERVAL", "MIDSCENE_MODEL_REASONING_EFFORT", "MIDSCENE_MODEL_REASONING_ENABLED", "MIDSCENE_MODEL_REASONING_BUDGET", "MIDSCENE_USE_VLM_UI_TARS", "MIDSCENE_USE_QWEN_VL", "MIDSCENE_USE_QWEN3_VL", "MIDSCENE_USE_DOUBAO_VISION", "MIDSCENE_USE_GEMINI", "MIDSCENE_USE_VL_MODEL", "OPENAI_API_KEY", "OPENAI_BASE_URL", "MIDSCENE_OPENAI_INIT_CONFIG_JSON", "MIDSCENE_OPENAI_HTTP_PROXY", "MIDSCENE_OPENAI_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_NAME", "MIDSCENE_INSIGHT_MODEL_SOCKS_PROXY", "MIDSCENE_INSIGHT_MODEL_HTTP_PROXY", "MIDSCENE_INSIGHT_MODEL_BASE_URL", "MIDSCENE_INSIGHT_MODEL_API_KEY", "MIDSCENE_INSIGHT_MODEL_INIT_CONFIG_JSON", "MIDSCENE_INSIGHT_MODEL_EXTRA_BODY_JSON", "MIDSCENE_INSIGHT_MODEL_TIMEOUT", "MIDSCENE_INSIGHT_MODEL_TEMPERATURE", "MIDSCENE_INSIGHT_MODEL_RETRY_COUNT", "MIDSCENE_INSIGHT_MODEL_RETRY_INTERVAL", "MIDSCENE_INSIGHT_MODEL_FAMILY", "MIDSCENE_INSIGHT_MODEL_REASONING_EFFORT", "MIDSCENE_INSIGHT_MODEL_REASONING_ENABLED", "MIDSCENE_INSIGHT_MODEL_REASONING_BUDGET", "MIDSCENE_PLANNING_MODEL_NAME", "MIDSCENE_PLANNING_MODEL_SOCKS_PROXY", "MIDSCENE_PLANNING_MODEL_HTTP_PROXY", "MIDSCENE_PLANNING_MODEL_BASE_URL", "MIDSCENE_PLANNING_MODEL_API_KEY", "MIDSCENE_PLANNING_MODEL_INIT_CONFIG_JSON", "MIDSCENE_PLANNING_MODEL_EXTRA_BODY_JSON", "MIDSCENE_PLANNING_MODEL_TIMEOUT", "MIDSCENE_PLANNING_MODEL_TEMPERATURE", "MIDSCENE_PLANNING_MODEL_RETRY_COUNT", "MIDSCENE_PLANNING_MODEL_RETRY_INTERVAL", "MIDSCENE_PLANNING_MODEL_FAMILY", "MIDSCENE_PLANNING_MODEL_REASONING_EFFORT", "MIDSCENE_PLANNING_MODEL_REASONING_ENABLED", "MIDSCENE_PLANNING_MODEL_REASONING_BUDGET", "MIDSCENE_MODEL_FAMILY"];
 export type TEnvKeys = (typeof ALL_ENV_KEYS)[number];
 export type TGlobalConfig = Record<TEnvKeys, string | undefined>;
 /**

package/dist/types/key-alias-utils.d.ts

@@ -1,7 +1,7 @@
 /**
- * Internal-only helpers for CLI/MCP argument key aliasing.
+ * Internal-only helpers for CLI argument key aliasing.
  * Not re-exported from the package entry point — keep consumers within
- * `cli/` and `mcp/`.
+ * `cli/`.
  */
 export declare function kebabToCamel(str: string): string;
 export declare function camelToKebab(str: string): string;

package/dist/types/utils.d.ts

@@ -11,7 +11,6 @@ export declare function generateHashId(rect: any, content?: string): string;
  * @throws Error with the provided message if the condition is false
  */
 export declare function assert(condition: any, message?: string): asserts condition;
-export declare function setIsMcp(value: boolean): void;
 export declare function logMsg(...message: Parameters<typeof console.log>): void;
 export declare function repeat(times: number, fn: (index: number) => Promise<void>): Promise<void>;
 export declare const escapeScriptTag: (html: string) => string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@midscene/shared",
-  "version": "1.9.8",
+  "version": "1.10.1-beta-20260624112700.0",
   "repository": "https://github.com/web-infra-dev/midscene",
   "homepage": "https://midscenejs.com/",
   "types": "./dist/types/index.d.ts",
@@ -57,16 +57,26 @@
       "import": "./dist/es/common.mjs",
       "require": "./dist/lib/common.js"
     },
-    "./mcp": {
-      "types": "./dist/types/mcp/index.d.ts",
-      "import": "./dist/es/mcp/index.mjs",
-      "require": "./dist/lib/mcp/index.js"
+    "./agent-tools": {
+      "types": "./dist/types/agent-tools/index.d.ts",
+      "import": "./dist/es/agent-tools/index.mjs",
+      "require": "./dist/lib/agent-tools/index.js"
+    },
+    "./agent-tools/*": {
+      "types": "./dist/types/agent-tools/*.d.ts",
+      "import": "./dist/es/agent-tools/*.mjs",
+      "require": "./dist/lib/agent-tools/*.js"
     },
     "./cli": {
       "types": "./dist/types/cli/index.d.ts",
       "import": "./dist/es/cli/index.mjs",
       "require": "./dist/lib/cli/index.js"
     },
+    "./recorder": {
+      "types": "./dist/types/recorder.d.ts",
+      "import": "./dist/es/recorder.mjs",
+      "require": "./dist/lib/recorder.js"
+    },
     "./logger": {
       "types": "./dist/types/logger.d.ts",
       "import": "./dist/es/logger.mjs",
@@ -84,11 +94,9 @@
     "README.md"
   ],
   "dependencies": {
-    "@modelcontextprotocol/sdk": "1.10.2",
     "@silvia-odwyer/photon": "0.3.3",
     "@silvia-odwyer/photon-node": "0.3.3",
     "debug": "4.4.0",
-    "express": "^4.21.2",
     "js-sha256": "0.11.0",
     "sharp": "^0.34.3",
     "dotenv": "^16.4.5",
@@ -98,11 +106,10 @@
   "devDependencies": {
     "@rslib/core": "^0.18.3",
     "@types/debug": "4.1.12",
-    "@types/express": "^4.17.21",
     "@types/node": "^18.0.0",
     "@ui-tars/shared": "1.2.0",
     "openai": "6.3.0",
-    "rimraf": "~3.0.2",
+    "rimraf": "~5.0.10",
     "typescript": "^5.8.3",
     "vitest": "3.0.5"
   },

package/src/{mcp → agent-tools}/base-tools.ts

@@ -1,6 +1,5 @@
 import { parseBase64 } from '@midscene/shared/img';
 import { getDebug } from '@midscene/shared/logger';
-import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
 import type { z } from 'zod';
 import { camelToKebab, getKeyAliases } from '../key-alias-utils';
 import {
@@ -29,7 +28,7 @@ import type {
   ToolSchema,
 } from './types';
 
-const debug = getDebug('mcp:base-tools');
+const debug = getDebug('agent-tools:base-tools');
 
 /**
  * Declarative description of a platform's agent init args.
@@ -39,11 +38,11 @@ const debug = getDebug('mcp:base-tools');
 export interface InitArgSpec<TInitParam> {
   /** Arg namespace, e.g. `android`, `ios`. */
   namespace: string;
-  /** Zod shape describing the init args. Field names drive the MCP schema. */
+  /** Zod shape describing the init args. Field names drive the tool schema. */
   shape: Record<string, z.ZodTypeAny>;
   /**
    * Optional CLI presentation hints. These affect `--help` output for
-   * single-platform CLIs but do not alter MCP/YAML protocol keys.
+   * single-platform CLIs but do not alter YAML protocol keys.
    */
   cli?: {
     /** Prefer bare `--device-id`-style options in platform CLI help output. */
@@ -61,7 +60,7 @@ export interface InitArgSpec<TInitParam> {
 }
 
 /**
- * Base class for platform-specific MCP tools.
+ * Base class for platform-specific Midscene tools.
  * @typeParam TAgent - Platform-specific agent type.
  * @typeParam TInitParam - Platform-specific init parameter consumed by
  *   `ensureAgent`. Defaults to `undefined` for platforms that take no args.
@@ -71,20 +70,19 @@ export abstract class BaseMidsceneTools<
   TInitParam = unknown,
 > implements IMidsceneTools
 {
-  protected mcpServer?: McpServer;
   protected agent?: TAgent;
   protected toolDefinitions: ToolDefinition[] = [];
 
   /**
    * Default options injected into every generated tool call (e.g. forced deep
-   * locate / deep think). Set from server/CLI behavior flags before
+   * locate / deep think). Set from startup/CLI behavior flags before
    * `initTools()` so they are baked into the generated tool handlers.
    * See https://github.com/web-infra-dev/midscene/issues/2446.
    */
   protected toolDefaults: ToolDefaults = {};
 
   /**
-   * Declarative init-arg spec. Subclasses that accept CLI/MCP init args should
+   * Declarative init-arg spec. Subclasses that accept CLI init args should
    * set this once and get `extractAgentInitParam` / `sanitizeToolArgs` /
    * `getAgentInitArgSchema` auto-implemented.
    *
@@ -108,7 +106,7 @@ export abstract class BaseMidsceneTools<
   }
 
   /**
-   * Extract a platform-specific agent init parameter from CLI/MCP tool args.
+   * Extract a platform-specific agent init parameter from CLI tool args.
    */
   protected extractAgentInitParam(
     args: Record<string, unknown>,
@@ -161,7 +159,7 @@ export abstract class BaseMidsceneTools<
    * show ergonomic bare flags while the underlying schema stays namespaced.
    * When `preferBareKeys` is enabled, single-platform CLIs only accept the
    * bare spellings; namespaced dotted spellings remain available through the
-   * MCP/YAML schema instead of the platform CLI surface.
+   * YAML schema instead of the platform CLI surface.
    */
   protected getAgentInitArgCliMetadata(): ToolCliMetadata | undefined {
     if (!this.initArgSpec?.cli) {
@@ -272,7 +270,6 @@ export abstract class BaseMidsceneTools<
     this.toolDefinitions.push(...platformTools);
 
     // 2. Get action space: use pre-set agent if available, otherwise temp device.
-    //    When called via mcpKitForAgent(), agent is set before initTools().
     //    For CLI usage, agent is deferred to the first real command.
     let actionSpace: ActionSpaceItem[];
     if (this.agent) {
@@ -313,28 +310,6 @@ export abstract class BaseMidsceneTools<
     debug('Total tools prepared:', this.toolDefinitions.length);
   }
 
-  /**
-   * Attach to MCP server and register all tools
-   */
-  public attachToServer(server: McpServer): void {
-    this.mcpServer = server;
-
-    if (this.toolDefinitions.length === 0) {
-      debug('Warning: No tools to register. Tools may be initialized lazily.');
-    }
-
-    for (const toolDef of this.toolDefinitions) {
-      this.mcpServer.tool(
-        toolDef.name,
-        toolDef.description,
-        toolDef.schema,
-        toolDef.handler,
-      );
-    }
-
-    debug('Registered', this.toolDefinitions.length, 'tools');
-  }
-
   /**
    * Cleanup method - destroy agent and release resources
    */

package/src/{mcp → agent-tools}/chrome-path.ts

@@ -1,5 +1,13 @@
 import { existsSync } from 'node:fs';
-import { MIDSCENE_MCP_CHROME_PATH, globalConfigManager } from '../env';
+import {
+  MIDSCENE_CHROME_PATH,
+  MIDSCENE_MCP_CHROME_PATH,
+  globalConfigManager,
+} from '../env';
+import { getDebug } from '../logger';
+
+const warnChromePath = getDebug('agent-tools:chrome-path', { console: true });
+let hasWarnedLegacyChromePath = false;
 
 export function getSystemChromePath(): string | undefined {
   const platform = process.platform;
@@ -33,9 +41,18 @@ export function getSystemChromePath(): string | undefined {
 }
 
 export function resolveChromePath(): string {
-  const envPath = globalConfigManager.getEnvConfigValue(
+  const primaryEnvPath =
+    globalConfigManager.getEnvConfigValue(MIDSCENE_CHROME_PATH);
+  const legacyEnvPath = globalConfigManager.getEnvConfigValue(
     MIDSCENE_MCP_CHROME_PATH,
   );
+  const envPath = primaryEnvPath || legacyEnvPath;
+  if (!primaryEnvPath && legacyEnvPath && !hasWarnedLegacyChromePath) {
+    warnChromePath(
+      'MIDSCENE_MCP_CHROME_PATH is deprecated. Use MIDSCENE_CHROME_PATH instead.',
+    );
+    hasWarnedLegacyChromePath = true;
+  }
   if (envPath && envPath !== 'auto' && existsSync(envPath)) {
     return envPath;
   }
@@ -43,6 +60,6 @@ export function resolveChromePath(): string {
   if (systemPath) return systemPath;
 
   throw new Error(
-    'Chrome not found. Install Google Chrome or set MIDSCENE_MCP_CHROME_PATH environment variable.',
+    'Chrome not found. Install Google Chrome or set MIDSCENE_CHROME_PATH environment variable.',
   );
 }

package/src/{mcp → agent-tools}/index.ts

@@ -1,4 +1,3 @@
-export * from './base-server';
 export * from './base-tools';
 export * from './tool-defaults';
 export * from './agent-behavior-init-args';
@@ -6,6 +5,4 @@ export * from './init-arg-utils';
 export * from './error-formatter';
 export * from './tool-generator';
 export * from './types';
-export * from './inject-report-html-plugin';
-export * from './launcher-helper';
 export * from './chrome-path';

package/src/{mcp → agent-tools}/init-arg-utils.ts

@@ -88,11 +88,11 @@ export function sanitizeNamespacedArgs(
 }
 
 /**
- * Build a flat MCP tool schema whose keys are dotted `"<namespace>.<field>"`.
+ * Build a flat tool schema whose keys are dotted `"<namespace>.<field>"`.
  *
  * We intentionally stay flat (rather than `{ namespace: z.object({...}) }`) so
- * that CLI (`--android.device-id`), MCP clients, and `--help` output all share
- * the same spelling. `readNamespacedArg` understands all three input shapes:
+ * that CLI (`--android.device-id`) and `--help` output share the same spelling.
+ * `readNamespacedArg` understands all three input shapes:
  * nested namespace object, dotted flat key, and bare key fallback.
  */
 export function createNamespacedInitArgSchema(

package/src/{mcp → agent-tools}/tool-defaults.ts

@@ -1,10 +1,10 @@
 /**
  * Unified, declarative mechanism for "force a default option on every tool
- * call" behaviors exposed by MCP servers and the device / Agent Skill CLIs.
+ * call" behaviors exposed by device and Agent Skill CLIs.
  *
  * Adding a new behavior flag (e.g. `--deep-search`) is a one-line change to
  * {@link TOOL_BEHAVIOR_FLAGS}: declare which default-option "bag" it fills.
- * The tool generator, servers, tools managers and CLI parsing are all generic
+ * The tool generator, tools managers and CLI parsing are all generic
  * over {@link ToolDefaults} and never need to learn about individual flags.
  *
  * See https://github.com/web-infra-dev/midscene/issues/2446.
@@ -91,13 +91,12 @@ export function resolveToolDefaults(
  *
  * Behavior flags (e.g. `--deep-locate`) are global: they may appear anywhere
  * in argv and are not tied to a specific sub-command. They are recognized by
- * exact kebab-case match — the same surface the MCP `parseArgs` config exposes
- * — and removed so a strict per-command parser never sees them. Every other
+ * exact kebab-case match and removed so a strict per-command parser never sees them. Every other
  * token is returned untouched and in order for that per-command parser.
  *
  * This is the single place that knows how a behavior flag looks on the command
- * line; both the device / Agent Skill CLI and the MCP launch path resolve their
- * defaults from {@link TOOL_BEHAVIOR_FLAGS} through here / {@link resolveToolDefaults}.
+ * line; the device / Agent Skill CLI resolves defaults from
+ * {@link TOOL_BEHAVIOR_FLAGS} through here / {@link resolveToolDefaults}.
  */
 export function stripBehaviorFlags(argv: readonly string[]): {
   rawArgs: string[];

package/src/{mcp → agent-tools}/tool-generator.ts

@@ -21,10 +21,10 @@ import { composeUserPrompt, promptInputExtraSchema } from './user-prompt';
 export { composeUserPrompt };
 
 /**
- * Generate MCP tool description from ActionSpaceItem
+ * Generate tool description from ActionSpaceItem.
  * Format: "actionName action, description. Parameters: param1 (type) - desc; param2 (type) - desc"
  */
-function describeActionForMCP(action: ActionSpaceItem): string {
+function describeActionForTool(action: ActionSpaceItem): string {
   const actionDesc = action.description || `Execute ${action.name} action`;
 
   if (!action.paramSchema) {
@@ -166,7 +166,7 @@ function transformSchemaField(
 /**
  * Extract and transform schema from action's paramSchema.
  *
- * CLI and MCP both expose parameters as named fields, so the only schema
+ * CLI tools expose parameters as named fields, so the only schema
  * shapes we can surface are ZodObject (any number of fields) or undefined
  * (the action takes no parameters). A primitive schema like `z.string()`
  * silently degraded to leaking the ZodString instance's prototype methods
@@ -188,7 +188,7 @@ function extractActionSchema(
       (paramSchema as unknown as { _def?: { typeName?: string } })?._def
         ?.typeName ?? 'unknown';
     throw new Error(
-      `Action "${actionName}" declared a non-object paramSchema (${typeName}). CLI and MCP tool schemas must be a ZodObject (e.g. z.object({ uri: z.string() })) or undefined. Wrap primitive fields in an object schema.`,
+      `Action "${actionName}" declared a non-object paramSchema (${typeName}). CLI tool schemas must be a ZodObject (e.g. z.object({ uri: z.string() })) or undefined. Wrap primitive fields in an object schema.`,
     );
   }
 
@@ -543,7 +543,7 @@ function mergeToolCliMetadata(
 }
 
 /**
- * Converts DeviceAction from actionSpace into MCP ToolDefinition
+ * Converts DeviceAction from actionSpace into ToolDefinition.
  * This is the core logic that removes need for hardcoded tool definitions
  */
 export function generateToolsFromActionSpace(
@@ -564,7 +564,7 @@ export function generateToolsFromActionSpace(
 
     return {
       name: action.name,
-      description: describeActionForMCP(action),
+      description: describeActionForTool(action),
       schema,
       cli: initArgCliMetadata,
       handler: async (args: Record<string, unknown>) => {

package/src/{mcp → agent-tools}/types.ts

@@ -1,4 +1,3 @@
-import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
 import type { z } from 'zod';
 import type { ToolDefaults } from './tool-defaults';
 
@@ -12,7 +11,7 @@ export const defaultAppLoadingTimeoutMs = 10000;
 export const defaultAppLoadingCheckIntervalMs = 2000;
 
 /**
- * Content item types for tool results (MCP compatible)
+ * Content item types for tool results.
  */
 export type ToolResultContent =
   | { type: 'text'; text: string }
@@ -26,7 +25,7 @@ export type ToolResultContent =
     };
 
 /**
- * Result type for tool execution (MCP compatible)
+ * Result type for tool execution.
  */
 export interface ToolResult {
   [x: string]: unknown;
@@ -58,7 +57,7 @@ export interface ToolCliMetadata {
 }
 
 /**
- * Tool definition for MCP server
+ * Tool definition for Midscene CLI and Skill surfaces.
  */
 export interface ToolDefinition<T = Record<string, unknown>> {
   name: string;
@@ -68,9 +67,6 @@ export interface ToolDefinition<T = Record<string, unknown>> {
   cli?: ToolCliMetadata;
 }
 
-/**
- * Tool type for mcpKitForAgent return value
- */
 export type Tool = ToolDefinition;
 
 /**
@@ -160,10 +156,9 @@ export interface BaseDevice {
 }
 
 /**
- * Interface for platform-specific MCP tools manager
+ * Interface for platform-specific tools manager.
  */
 export interface IMidsceneTools {
-  attachToServer(server: McpServer): void;
   initTools(): Promise<void>;
   destroy?(): Promise<void>;
   setToolDefaults?(toolDefaults: ToolDefaults): void;

package/src/cli/cli-args.ts

@@ -1,6 +1,6 @@
 import { z } from 'zod';
+import type { ToolCliOption, ToolDefinition } from '../agent-tools/types';
 import { getKeyAliases } from '../key-alias-utils';
-import type { ToolCliOption, ToolDefinition } from '../mcp/types';
 import { CLIError } from './cli-error';
 
 export function parseValue(raw: string): unknown {

package/src/cli/cli-runner.ts

@@ -2,14 +2,14 @@ import { existsSync, writeFileSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 import dotenv from 'dotenv';
-import { getDebug } from '../logger';
-import type { BaseMidsceneTools } from '../mcp/base-tools';
-import { stripBehaviorFlags } from '../mcp/tool-defaults';
+import type { BaseMidsceneTools } from '../agent-tools/base-tools';
+import { stripBehaviorFlags } from '../agent-tools/tool-defaults';
 import type {
   ToolDefinition,
   ToolResult,
   ToolResultContent,
-} from '../mcp/types';
+} from '../agent-tools/types';
+import { getDebug } from '../logger';
 import {
   canonicalizeCliArgKeys,
   formatCliValidationError,

package/src/env/types.ts

@@ -8,10 +8,11 @@ export const MIDSCENE_DEBUG_MODEL_RESPONSE = 'MIDSCENE_DEBUG_MODEL_RESPONSE';
 export const MIDSCENE_DANGEROUSLY_PRINT_ALL_CONFIG =
   'MIDSCENE_DANGEROUSLY_PRINT_ALL_CONFIG';
 export const MIDSCENE_DEBUG_MODE = 'MIDSCENE_DEBUG_MODE';
-export const MIDSCENE_MCP_USE_PUPPETEER_MODE =
-  'MIDSCENE_MCP_USE_PUPPETEER_MODE';
+export const MIDSCENE_CHROME_PATH = 'MIDSCENE_CHROME_PATH';
+/**
+ * @deprecated Use MIDSCENE_CHROME_PATH instead. This is kept for backward compatibility.
+ */
 export const MIDSCENE_MCP_CHROME_PATH = 'MIDSCENE_MCP_CHROME_PATH';
-export const MIDSCENE_MCP_ANDROID_MODE = 'MIDSCENE_MCP_ANDROID_MODE';
 export const DOCKER_CONTAINER = 'DOCKER_CONTAINER';
 
 // Observability
@@ -166,8 +167,6 @@ export const BASIC_ENV_KEYS = [
 
 export const BOOLEAN_ENV_KEYS = [
   MIDSCENE_CACHE,
-  MIDSCENE_MCP_USE_PUPPETEER_MODE,
-  MIDSCENE_MCP_ANDROID_MODE,
   MIDSCENE_LANGSMITH_DEBUG,
   MIDSCENE_LANGFUSE_DEBUG,
   MIDSCENE_REPORT_QUIET,
@@ -188,6 +187,7 @@ export const STRING_ENV_KEYS = [
   MIDSCENE_REPORT_TAG_NAME,
   MIDSCENE_PREFERRED_LANGUAGE,
   MATCH_BY_POSITION,
+  MIDSCENE_CHROME_PATH,
   MIDSCENE_MCP_CHROME_PATH,
   DOCKER_CONTAINER,
 ] as const;

package/src/key-alias-utils.ts

@@ -1,7 +1,7 @@
 /**
- * Internal-only helpers for CLI/MCP argument key aliasing.
+ * Internal-only helpers for CLI argument key aliasing.
  * Not re-exported from the package entry point — keep consumers within
- * `cli/` and `mcp/`.
+ * `cli/`.
  */
 
 export function kebabToCamel(str: string): string {