viyv-browser-mcp 0.7.3 → 0.7.5

package/dist/index.js

@@ -75,7 +75,7 @@ var BRIDGE = {
   /** Loopback address for bridge */
   TCP_HOST: "127.0.0.1",
   /** Maximum simultaneous MCP Server connections */
-  MAX_CONNECTIONS: 10,
+  MAX_CONNECTIONS: 100,
   /** TTL for request routing entries (ms) */
   REQUEST_ROUTING_TTL: 6e4
 };
@@ -124,10 +124,10 @@ function syncNativeHostBinary() {
   }
 }
 function shouldUpdateBridge(bridgeVersion) {
-  return compareSemver("0.7.3", bridgeVersion) > 0;
+  return compareSemver("0.7.5", bridgeVersion) > 0;
 }
 function getPackageVersion() {
-  return "0.7.3";
+  return "0.7.5";
 }
 function compareSemver(a, b) {
   const pa = a.split(".").map(Number);
@@ -317,6 +317,7 @@ var LOG_PREFIX2 = "[viyv-browser:native-host]";
 var ROUTING_CLEANUP_INTERVAL = 3e4;
 function startBridge(options) {
   const { port = BRIDGE.TCP_PORT, host = BRIDGE.TCP_HOST, onError } = options;
+  const maxConnections = Number(process.env.VIYV_BRIDGE_MAX_CONNECTIONS) || BRIDGE.MAX_CONNECTIONS;
   const mcpConnections = /* @__PURE__ */ new Map();
   const requestOrigin = /* @__PURE__ */ new Map();
   const agentToConn = /* @__PURE__ */ new Map();
@@ -735,9 +736,9 @@ function startBridge(options) {
     });
   }
   function handleMcpConnection(socket, initialData) {
-    if (mcpConnections.size >= BRIDGE.MAX_CONNECTIONS) {
+    if (mcpConnections.size >= maxConnections) {
       process.stderr.write(
-        `${LOG_PREFIX2} Max connections (${BRIDGE.MAX_CONNECTIONS}) reached, rejecting
+        `${LOG_PREFIX2} Max connections (${maxConnections}) reached, rejecting
 `
       );
       socket.destroy();
@@ -1024,8 +1025,15 @@ var TOPICS = {
    - Don't use for: checking text values, verifying form state
    - Do use for: layout verification, image content, charts
 
+6. indexed_page \u2192 Alternative text-based DOM representation with numbered interactive elements.
+   - Returns [0]<button>Submit</>, [1]<input placeholder=Search/>, etc.
+   - Use indexed_action(snapshotId, index, action) to interact by number
+   - No screenshots needed \u2014 works with text-only LLMs, lower token cost
+   - snapshotId expires on page change \u2014 call indexed_page again if stale
+
 Recommended workflow:
-  page_outline \u2192 read_page(section) \u2192 find(specific element) \u2192 click/form_input`
+  page_outline \u2192 read_page(section) \u2192 find(specific element) \u2192 click/form_input
+  OR: indexed_page \u2192 indexed_action(click/type/select_option) for lower-cost automation`
   },
   interaction: {
     title: "Interaction \u2014 How to interact with elements",
@@ -1073,6 +1081,14 @@ file_upload \u2014 For uploading files or dropping files onto elements
   - Don't click file inputs \u2014 use file_upload instead
   - Best for: file inputs, canvas apps (Canva, Figma), custom drop zones
 
+indexed_action \u2014 Index-based interaction (alternative to ref-based tools above)
+  - Requires snapshotId from indexed_page
+  - Actions: click, type, select_option, hover, scroll_to
+  - click/hover: CDP dispatch (trusted events, SPA compatible)
+  - type: CDP click to focus + content script value setting
+  - select_option: content script DOM manipulation
+  - Best for: batch automation with text-only LLMs, lower token cost
+
 Common patterns:
   Single field: find(input) \u2192 form_input(ref, value, submit: true)
   Multi-field form: read_page \u2192 form_fill([{ref, value}, ...], submit: true)
@@ -1081,7 +1097,8 @@ Common patterns:
   Dropdown: click(select ref) \u2192 key("ArrowDown ArrowDown Enter")
   Search: form_input(search input ref, query, submit: true)
   File upload: find(file input) \u2192 file_upload(ref, paths)
-  File drop: find(canvas) \u2192 file_upload(ref, paths) or file_upload(coordinate, paths)`
+  File drop: find(canvas) \u2192 file_upload(ref, paths) or file_upload(coordinate, paths)
+  Index-based: indexed_page \u2192 indexed_action(click, index=N) \u2192 wait_for(navigation)`
   },
   navigation: {
     title: "Navigation \u2014 How to navigate between pages",
@@ -1512,7 +1529,7 @@ var CLICK_RETURNS = `{
   clicked: true
   newTab?: { tabId, url } // only if target="_blank" link detected
 }`;
-var CLICK_RELATED = ["form_input", "hover", "drag", "javascript_exec"];
+var CLICK_RELATED = ["form_input", "hover", "drag", "javascript_exec", "indexed_action"];
 
 // src/tools/core/drag.ts
 var DRAG_DESCRIPTION = `Drag from start coordinate to end coordinate.
@@ -1608,7 +1625,7 @@ var FORM_INPUT_RETURNS = `{
   navigated?: boolean   // if page navigated after submit
   url?: string          // new URL if navigated
 }`;
-var FORM_INPUT_RELATED = ["click", "type", "find", "read_page"];
+var FORM_INPUT_RELATED = ["click", "type", "find", "read_page", "indexed_action"];
 
 // src/tools/core/get-page-text.ts
 var GET_PAGE_TEXT_DESCRIPTION = `Extract readable text from page. Use query for semantic filtering.
@@ -1628,6 +1645,66 @@ var GET_PAGE_TEXT_RETURNS = `// Without query:
 { sections: [{ index, tag, role, ariaLabel, heading, text, truncated }], truncated: boolean, charCount: number }`;
 var GET_PAGE_TEXT_RELATED = ["read_page", "find", "screenshot"];
 
+// src/tools/core/indexed-action.ts
+var INDEXED_ACTION_DESCRIPTION = `Interact with an element by its [index] from indexed_page. Actions: click, type, select_option, hover, scroll_to.
+Returns: { success, action, index }`;
+var INDEXED_ACTION_DETAIL = `Act on an indexed element from indexed_page output.
+
+Actions:
+- click: Click the element. Supports modifier keys (ctrl, shift, alt, meta).
+- type: Set text value on input/textarea/contenteditable. Clears existing content.
+- select_option: Select a dropdown option by text match.
+- hover: Hover over the element.
+- scroll_to: Scroll the element into view.
+
+The snapshotId from indexed_page is required. If the page has changed since indexed_page was called, this tool returns an error \u2014 call indexed_page again to get a fresh snapshot.
+
+Examples:
+  indexed_action(tabId=1, snapshotId=3, index=2, action="click")
+  indexed_action(tabId=1, snapshotId=3, index=1, action="type", text="hello")
+  indexed_action(tabId=1, snapshotId=3, index=5, action="select_option", option="Option A")`;
+var INDEXED_ACTION_RETURNS = `{
+  action: string       // the action performed
+  index: number        // the element index
+  success: boolean     // whether the action succeeded
+  value?: string       // set value (for type/select_option)
+  tagName?: string     // element tag name
+}`;
+var INDEXED_ACTION_RELATED = ["indexed_page", "click", "form_input"];
+
+// src/tools/core/indexed-page.ts
+var INDEXED_PAGE_DESCRIPTION = `Get page DOM as indexed text. Interactive elements marked [0], [1], etc. Use indexed_action to act by index.
+Returns: { tree, snapshotId, elementCount, interactiveCount }`;
+var INDEXED_PAGE_DETAIL = `Alternative to read_page that returns a text-based DOM representation with numbered interactive elements.
+
+Format example:
+  [0]<a aria-label=Home>Home />
+  [1]<input placeholder=Search />
+  [2]<button>Submit />
+  Some text content
+  [3]<a>Learn more />
+
+Each [N] marks an interactive element. Use indexed_action with the index number to click, type, or interact.
+
+Options:
+- viewportOnly: only include elements visible in the viewport (default: false = full page)
+- maxChars: limit output size (default 50000)
+- includeAttributes: additional HTML attributes to include in output
+
+The returned snapshotId must be passed to indexed_action. It expires after page changes \u2014 call indexed_page again if indexed_action reports a stale snapshot.
+
+Advantages over read_page:
+- Lower token cost (no screenshots needed)
+- Works with text-only LLMs
+- Direct index-based interaction without ref resolution`;
+var INDEXED_PAGE_RETURNS = `{
+  tree: string            // indexed DOM text
+  snapshotId: number      // pass to indexed_action
+  elementCount: number    // total elements
+  interactiveCount: number // indexed interactive elements
+}`;
+var INDEXED_PAGE_RELATED = ["indexed_action", "read_page", "screenshot"];
+
 // src/tools/core/handle-dialog.ts
 var HANDLE_DIALOG_DESCRIPTION = `Handle JS dialog (alert/confirm/prompt) or set next auto-handle policy.
 Returns: { handled } or { policy_set }`;
@@ -1762,7 +1839,7 @@ var PAGE_OUTLINE_RETURNS = `{
   }]
   note?: string     // only if no landmarks found
 }`;
-var PAGE_OUTLINE_RELATED = ["read_page", "find", "get_page_text"];
+var PAGE_OUTLINE_RELATED = ["read_page", "find", "get_page_text", "indexed_page"];
 
 // src/tools/core/read-page.ts
 var READ_PAGE_DESCRIPTION = `Get page accessibility tree with element refs for interaction. Call page_outline first on unfamiliar pages.
@@ -1787,7 +1864,7 @@ var READ_PAGE_RETURNS = `{
   elementCount: number // total elements in tree
   truncated: boolean   // whether output hit maxChars limit
 }`;
-var READ_PAGE_RELATED = ["page_outline", "find", "get_page_text", "screenshot"];
+var READ_PAGE_RELATED = ["page_outline", "find", "get_page_text", "screenshot", "indexed_page"];
 
 // src/tools/core/read-table.ts
 var READ_TABLE_DESCRIPTION = `Extract HTML table data from web page by ref. Supports <table> and ARIA role="table".
@@ -3436,6 +3513,37 @@ var readPageTool = {
     )
   })
 };
+var indexedPageTool = {
+  name: "indexed_page",
+  description: INDEXED_PAGE_DESCRIPTION,
+  detail: INDEXED_PAGE_DETAIL,
+  returns: INDEXED_PAGE_RETURNS,
+  category: "core",
+  related: INDEXED_PAGE_RELATED,
+  inputSchema: z.object({
+    tabId: z.coerce.number().describe("Tab ID"),
+    viewportOnly: z.boolean().optional().describe("Only include viewport-visible elements (default: false)"),
+    maxChars: z.coerce.number().optional().describe("Max output characters (default: 50000)"),
+    includeAttributes: z.array(z.string()).optional().describe("Additional HTML attributes to include in output")
+  })
+};
+var indexedActionTool = {
+  name: "indexed_action",
+  description: INDEXED_ACTION_DESCRIPTION,
+  detail: INDEXED_ACTION_DETAIL,
+  returns: INDEXED_ACTION_RETURNS,
+  category: "core",
+  related: INDEXED_ACTION_RELATED,
+  inputSchema: z.object({
+    tabId: z.coerce.number().describe("Tab ID"),
+    snapshotId: z.coerce.number().describe("Snapshot ID from indexed_page"),
+    index: z.coerce.number().min(0).describe("Element index from indexed_page output"),
+    action: z.enum(["click", "type", "select_option", "hover", "scroll_to"]).describe("Action to perform on the element"),
+    text: z.string().optional().describe('Text to type (required for action: "type")'),
+    option: z.string().optional().describe('Option text to select (required for action: "select_option")'),
+    modifiers: z.string().optional().describe('Modifier keys for click (e.g. "ctrl+shift")')
+  })
+};
 var findTool = {
   name: "find",
   description: FIND_DESCRIPTION,
@@ -4786,6 +4894,8 @@ var allTools = [
   hoverTool,
   dragTool,
   readPageTool,
+  indexedPageTool,
+  indexedActionTool,
   findTool,
   inspectTool,
   formInputTool,