zerochat 9.0.0__tar.gz → 9.1.0__tar.gz

{zerochat-9.0.0/zerochat.egg-info → zerochat-9.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: zerochat
-Version: 9.0.0
+Version: 9.1.0
 Summary: Private local web client for AI models, tools and MCP services
 Author: Alberto Alday
 License-Expression: MIT
@@ -22,11 +22,11 @@ Dynamic: license-file
 
 # ZeroChat
 
-ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **8.x**, que consolida las funcionalidades existentes. El desarrollo de nuevas funcionalidades queda en pausa por el momento.
+ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **9.x**. Es un proyecto personal, de libre uso con licencia MIT si a alguien le sirve; su objetivo actual es servir de frontend único para un LLM local (Swift-Qwen3.8) con una API compatible con OpenAI.
 
 La interfaz se sirve como archivos estáticos: `zerochat.html` necesita sus módulos `js/` y `css/`. El historial y la configuración se guardan en el navegador; no se sincronizan automáticamente entre dispositivos. Los proveedores remotos reciben el contenido incluido en las peticiones, y las herramientas web pueden utilizar servicios intermediarios. Almacenamiento local no significa privacidad absoluta.
 
-*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **8.x** series, which consolidates existing features. New feature development is paused for now.
+*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **9.x** series. It is a personal project, free to use under the MIT licence if it helps anyone; its current goal is to serve as the single frontend for a local LLM (Swift-Qwen3.8) through an OpenAI-compatible API.
 
 The interface is served as static files: `zerochat.html` needs its `js/` and `css/` modules. History and settings are stored in the browser, without automatic cross-device sync. Remote providers receive content included in requests, and web tools may use intermediary services. Local storage does not mean absolute privacy.
 
@@ -35,7 +35,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Perfiles:** OpenAI y APIs de chat compatibles, Ollama, OpenRouter, Claude y Gemini. La visión, las herramientas y el razonamiento dependen del modelo y del adaptador.
 - **WebLLM:** inferencia en el navegador con WebGPU, tras descargar un modelo. Necesita hardware compatible y memoria suficiente.
 - **Documentos:** adjuntos y colecciones RAG con extracción local y búsqueda léxica mediante Orama; sin embeddings. Los fragmentos recuperados se envían al modelo elegido. La extracción de PDF tiene límites y no incluye OCR general.
-- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, planes y puntos de control. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
+- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, preguntas al usuario y subagentes de solo lectura. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
 - **Historial y diagnóstico:** ramas de conversación, exportación JSON y Markdown, impresión a PDF, métricas de contexto y revisión de peticiones antes del envío.
 
 *English:*
@@ -43,7 +43,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Profiles:** OpenAI and compatible chat APIs, Ollama, OpenRouter, Claude and Gemini. Vision, tools and reasoning depend on the model and adapter.
 - **WebLLM:** browser inference through WebGPU after downloading a model. Requires compatible hardware and enough memory.
 - **Documents:** attachments and RAG collections with local extraction and Orama lexical search, without embeddings. Retrieved passages are sent to the selected model. PDF extraction has limitations and no general OCR.
-- **Tools:** isolated JavaScript, web search and retrieval, charts, plans and checkpoints. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
+- **Tools:** isolated JavaScript, web search and retrieval, charts, questions to the user and read-only subagents. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
 - **History and diagnostics:** conversation branches, JSON and Markdown export, printing to PDF, context metrics and request inspection before sending.
 
 ## Inicio rápido / Quick start
@@ -93,8 +93,8 @@ Deleting `~/zerochat/` removes that state and managed MCP services, not the exec
 
 ## Documentación y ayuda / Documentation and help
 
-- [Ayuda de ZeroChat 8.x — Español](https://albalday.github.io/zerochat/help/index.html)
-- [ZeroChat 8.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
+- [Ayuda de ZeroChat 9.x — Español](https://albalday.github.io/zerochat/help/index.html)
+- [ZeroChat 9.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
 
 Las guías explican configuración, uso y límites.
 

{zerochat-9.0.0 → zerochat-9.1.0}/README.md

@@ -1,10 +1,10 @@
 # ZeroChat
 
-ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **8.x**, que consolida las funcionalidades existentes. El desarrollo de nuevas funcionalidades queda en pausa por el momento.
+ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **9.x**. Es un proyecto personal, de libre uso con licencia MIT si a alguien le sirve; su objetivo actual es servir de frontend único para un LLM local (Swift-Qwen3.8) con una API compatible con OpenAI.
 
 La interfaz se sirve como archivos estáticos: `zerochat.html` necesita sus módulos `js/` y `css/`. El historial y la configuración se guardan en el navegador; no se sincronizan automáticamente entre dispositivos. Los proveedores remotos reciben el contenido incluido en las peticiones, y las herramientas web pueden utilizar servicios intermediarios. Almacenamiento local no significa privacidad absoluta.
 
-*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **8.x** series, which consolidates existing features. New feature development is paused for now.
+*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **9.x** series. It is a personal project, free to use under the MIT licence if it helps anyone; its current goal is to serve as the single frontend for a local LLM (Swift-Qwen3.8) through an OpenAI-compatible API.
 
 The interface is served as static files: `zerochat.html` needs its `js/` and `css/` modules. History and settings are stored in the browser, without automatic cross-device sync. Remote providers receive content included in requests, and web tools may use intermediary services. Local storage does not mean absolute privacy.
 
@@ -13,7 +13,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Perfiles:** OpenAI y APIs de chat compatibles, Ollama, OpenRouter, Claude y Gemini. La visión, las herramientas y el razonamiento dependen del modelo y del adaptador.
 - **WebLLM:** inferencia en el navegador con WebGPU, tras descargar un modelo. Necesita hardware compatible y memoria suficiente.
 - **Documentos:** adjuntos y colecciones RAG con extracción local y búsqueda léxica mediante Orama; sin embeddings. Los fragmentos recuperados se envían al modelo elegido. La extracción de PDF tiene límites y no incluye OCR general.
-- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, planes y puntos de control. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
+- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, preguntas al usuario y subagentes de solo lectura. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
 - **Historial y diagnóstico:** ramas de conversación, exportación JSON y Markdown, impresión a PDF, métricas de contexto y revisión de peticiones antes del envío.
 
 *English:*
@@ -21,7 +21,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Profiles:** OpenAI and compatible chat APIs, Ollama, OpenRouter, Claude and Gemini. Vision, tools and reasoning depend on the model and adapter.
 - **WebLLM:** browser inference through WebGPU after downloading a model. Requires compatible hardware and enough memory.
 - **Documents:** attachments and RAG collections with local extraction and Orama lexical search, without embeddings. Retrieved passages are sent to the selected model. PDF extraction has limitations and no general OCR.
-- **Tools:** isolated JavaScript, web search and retrieval, charts, plans and checkpoints. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
+- **Tools:** isolated JavaScript, web search and retrieval, charts, questions to the user and read-only subagents. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
 - **History and diagnostics:** conversation branches, JSON and Markdown export, printing to PDF, context metrics and request inspection before sending.
 
 ## Inicio rápido / Quick start
@@ -71,8 +71,8 @@ Deleting `~/zerochat/` removes that state and managed MCP services, not the exec
 
 ## Documentación y ayuda / Documentation and help
 
-- [Ayuda de ZeroChat 8.x — Español](https://albalday.github.io/zerochat/help/index.html)
-- [ZeroChat 8.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
+- [Ayuda de ZeroChat 9.x — Español](https://albalday.github.io/zerochat/help/index.html)
+- [ZeroChat 9.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
 
 Las guías explican configuración, uso y límites.
 

{zerochat-9.0.0 → zerochat-9.1.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "zerochat"
-version = "9.0.0"
+version = "9.1.0"
 description = "Private local web client for AI models, tools and MCP services"
 readme = "README.md"
 requires-python = ">=3.10"

{zerochat-9.0.0 → zerochat-9.1.0/zerochat.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: zerochat
-Version: 9.0.0
+Version: 9.1.0
 Summary: Private local web client for AI models, tools and MCP services
 Author: Alberto Alday
 License-Expression: MIT
@@ -22,11 +22,11 @@ Dynamic: license-file
 
 # ZeroChat
 
-ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **8.x**, que consolida las funcionalidades existentes. El desarrollo de nuevas funcionalidades queda en pausa por el momento.
+ZeroChat es un cliente web de uso personal para conversar con modelos de IA, consultar documentos y utilizar herramientas. Esta documentación corresponde a la serie **9.x**. Es un proyecto personal, de libre uso con licencia MIT si a alguien le sirve; su objetivo actual es servir de frontend único para un LLM local (Swift-Qwen3.8) con una API compatible con OpenAI.
 
 La interfaz se sirve como archivos estáticos: `zerochat.html` necesita sus módulos `js/` y `css/`. El historial y la configuración se guardan en el navegador; no se sincronizan automáticamente entre dispositivos. Los proveedores remotos reciben el contenido incluido en las peticiones, y las herramientas web pueden utilizar servicios intermediarios. Almacenamiento local no significa privacidad absoluta.
 
-*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **8.x** series, which consolidates existing features. New feature development is paused for now.
+*English:* ZeroChat is a web client for personal use: chatting with AI models, consulting documents and using tools. This documentation covers the **9.x** series. It is a personal project, free to use under the MIT licence if it helps anyone; its current goal is to serve as the single frontend for a local LLM (Swift-Qwen3.8) through an OpenAI-compatible API.
 
 The interface is served as static files: `zerochat.html` needs its `js/` and `css/` modules. History and settings are stored in the browser, without automatic cross-device sync. Remote providers receive content included in requests, and web tools may use intermediary services. Local storage does not mean absolute privacy.
 
@@ -35,7 +35,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Perfiles:** OpenAI y APIs de chat compatibles, Ollama, OpenRouter, Claude y Gemini. La visión, las herramientas y el razonamiento dependen del modelo y del adaptador.
 - **WebLLM:** inferencia en el navegador con WebGPU, tras descargar un modelo. Necesita hardware compatible y memoria suficiente.
 - **Documentos:** adjuntos y colecciones RAG con extracción local y búsqueda léxica mediante Orama; sin embeddings. Los fragmentos recuperados se envían al modelo elegido. La extracción de PDF tiene límites y no incluye OCR general.
-- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, planes y puntos de control. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
+- **Herramientas:** JavaScript aislado, búsqueda y lectura web, gráficos, preguntas al usuario y subagentes de solo lectura. El backend opcional añade archivos, comandos, diagnósticos, automatización de navegador y servicios MCP.
 - **Historial y diagnóstico:** ramas de conversación, exportación JSON y Markdown, impresión a PDF, métricas de contexto y revisión de peticiones antes del envío.
 
 *English:*
@@ -43,7 +43,7 @@ The interface is served as static files: `zerochat.html` needs its `js/` and `cs
 - **Profiles:** OpenAI and compatible chat APIs, Ollama, OpenRouter, Claude and Gemini. Vision, tools and reasoning depend on the model and adapter.
 - **WebLLM:** browser inference through WebGPU after downloading a model. Requires compatible hardware and enough memory.
 - **Documents:** attachments and RAG collections with local extraction and Orama lexical search, without embeddings. Retrieved passages are sent to the selected model. PDF extraction has limitations and no general OCR.
-- **Tools:** isolated JavaScript, web search and retrieval, charts, plans and checkpoints. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
+- **Tools:** isolated JavaScript, web search and retrieval, charts, questions to the user and read-only subagents. The optional backend adds files, commands, diagnostics, browser automation and MCP services.
 - **History and diagnostics:** conversation branches, JSON and Markdown export, printing to PDF, context metrics and request inspection before sending.
 
 ## Inicio rápido / Quick start
@@ -93,8 +93,8 @@ Deleting `~/zerochat/` removes that state and managed MCP services, not the exec
 
 ## Documentación y ayuda / Documentation and help
 
-- [Ayuda de ZeroChat 8.x — Español](https://albalday.github.io/zerochat/help/index.html)
-- [ZeroChat 8.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
+- [Ayuda de ZeroChat 9.x — Español](https://albalday.github.io/zerochat/help/index.html)
+- [ZeroChat 9.x help — English](https://albalday.github.io/zerochat/help/en/index.html)
 
 Las guías explican configuración, uso y límites.
 

{zerochat-9.0.0 → zerochat-9.1.0}/zerochat.py

@@ -35,6 +35,7 @@ import tempfile
 import threading
 import time
 import traceback
+import urllib.error
 import urllib.request
 from urllib.parse import urlencode
 import venv
@@ -42,7 +43,7 @@ import webbrowser
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
 
-SOURCE_BACKEND_VERSION = "9.0.0"
+SOURCE_BACKEND_VERSION = "9.1.0"
 
 def _read_source_version(filename: str) -> str | None:
     """Lee la versión de un archivo del repositorio cuando se ejecuta desde fuentes."""
@@ -53,7 +54,7 @@ def _read_source_version(filename: str) -> str | None:
             match = re.search(r'version\s*=\s*"([0-9]+\.[0-9]+\.[0-9]+)"', content)
             if match:
                 return match.group(1)
-    except Exception:
+    except (OSError, re.error):
         pass
     return None
 
@@ -77,7 +78,7 @@ def _read_ui_version() -> str:
             data = json.loads(pkg_path.read_text(encoding="utf-8"))
             if isinstance(data.get("version"), str):
                 return data["version"].strip()
-    except Exception:
+    except (OSError, json.JSONDecodeError):
         pass
     return BACKEND_PACKAGE_VERSION
 
@@ -294,13 +295,13 @@ def get_daily_token() -> str:
             data = json.loads(token_file.read_text(encoding="utf-8"))
             if data.get("date") == today and data.get("token") and isinstance(data["token"], str):
                 return data["token"]
-        except Exception:
+        except (OSError, json.JSONDecodeError):
             pass
     token = secrets.token_urlsafe(32)
     try:
         _write_private_text(token_file, json.dumps({"token": token, "date": today}, indent=2))
-    except Exception:
-        pass
+    except OSError as err:
+        log_event(f"[zerochat] Advertencia: no se pudo guardar el token diario en {token_file}: {err}")
     return token
 
 
@@ -402,8 +403,8 @@ def check_version():
             else:
                 notice += f" Actualiza con: curl -sSL {REMOTE_SCRIPT_URL} -o zerochat.py"
             add_notice(notice)
-    except Exception:
-        # Modo offline o timeout ignorado de forma segura
+    except (urllib.error.URLError, TimeoutError, OSError):
+        # Modo offline o sin conectividad al comprobar versiones remotas.
         pass
 # ==============================================================================
 # Ejecución de comandos y tareas en segundo plano
@@ -465,15 +466,18 @@ def _kill_process_tree(proc: subprocess.Popen):
                 creationflags=getattr(subprocess, "CREATE_NO_WINDOW", 0)
             )
             if proc.poll() is None:
-                proc.kill()
+                try:
+                    proc.kill()
+                except (ProcessLookupError, OSError, AttributeError):
+                    pass
         elif hasattr(os, "killpg") and hasattr(os, "getpgid"):
             os.killpg(os.getpgid(proc.pid), signal.SIGKILL)
         else:
             proc.kill()
-    except Exception:
+    except (ProcessLookupError, PermissionError, OSError, AttributeError):
         try:
             proc.kill()
-        except Exception:
+        except (ProcessLookupError, OSError, AttributeError):
             pass
 
 
@@ -1110,6 +1114,9 @@ def materialize_managed_services(services_root: Path | None = None) -> None:
 # Herramientas Locales Core
 # ==============================================================================
 
+MAX_HTTP_BODY_BYTES = 1024 * 1024
+MAX_PATH_LENGTH = 4096
+MAX_COMMAND_LENGTH = 16384
 BROWSER_ACTIONS = ("navigate", "screenshot", "click", "fill")
 
 
@@ -1272,8 +1279,8 @@ def write_file(path: str, content: str) -> str:
         return _tool_error(str(e))
 
 
-def edit_file(path: str, old_str: str, new_str: str) -> str:
-    """Replace the single occurrence of old_str with new_str in an existing file."""
+def edit_file(path: str, old_str: str, new_str: str, replace_all: bool = False) -> str:
+    """Replace old_str with new_str in an existing file: its single occurrence, or all of them with replace_all."""
     try:
         target = Path(path).expanduser().resolve()
         if not target.exists():
@@ -1284,19 +1291,21 @@ def edit_file(path: str, old_str: str, new_str: str) -> str:
         with open(target, "r", encoding="utf-8", errors="replace") as f:
             file_text = f.read()
 
+        if not old_str:
+            return _tool_error("old_str is empty. Provide the exact text to replace.")
         occurrences = file_text.count(old_str)
         if occurrences == 0:
             return _tool_error(f"Target text was not found in '{path}'. Check exact file content with read_file.")
-        if occurrences > 1:
-            return _tool_error(f"Found {occurrences} matches for the snippet in '{path}'. Provide more surrounding context in old_str to ensure a unique match.")
+        if occurrences > 1 and not replace_all:
+            return _tool_error(f"Found {occurrences} matches for the snippet in '{path}'. Provide more surrounding context in old_str to ensure a unique match, or set replace_all to replace every match.")
 
-        updated_text = file_text.replace(old_str, new_str, 1)
+        updated_text = file_text.replace(old_str, new_str) if replace_all else file_text.replace(old_str, new_str, 1)
         _atomic_write_text(target, updated_text)
         return json.dumps({
             "success": True,
             "path": str(target),
             "bytes_written": len(updated_text.encode("utf-8")),
-            "replacements": 1
+            "replacements": occurrences if replace_all else 1
         }, ensure_ascii=False, indent=2)
     except Exception as e:
         return _tool_error(str(e))
@@ -1445,7 +1454,7 @@ def get_diagnostics(path: str | None = None) -> str:
                                 "severity": "error",
                                 "message": msg
                             })
-                    except Exception:
+                    except (subprocess.SubprocessError, OSError):
                         pass
             elif ext == ".json":
                 try:
@@ -1676,9 +1685,9 @@ try {
                             self._process.stdin.write(json.dumps({"action": "close"}) + "\n")
                             self._process.stdin.flush()
                             self._process.wait(timeout=2)
-                        except Exception:
+                        except (OSError, subprocess.SubprocessError):
                             self._process.kill()
-                except Exception:
+                except (OSError, subprocess.SubprocessError):
                     pass
                 self._process = None
 
@@ -1746,7 +1755,7 @@ LOCAL_TOOLS_DEFINITIONS = [
         "inputSchema": {
             "type": "object",
             "properties": {
-                "path": {"type": "string", "description": "Relative or absolute directory path (defaults to '.')"},
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "Relative or absolute directory path (defaults to '.')"},
                 "recursive": {"type": "boolean", "description": "If true, traverses subdirectories up to depth 3", "default": False}
             }
         }
@@ -1757,7 +1766,7 @@ LOCAL_TOOLS_DEFINITIONS = [
         "inputSchema": {
             "type": "object",
             "properties": {
-                "path": {"type": "string", "description": "File path"},
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "File path"},
                 "start_line": {"type": "integer", "description": "First line (1-based)", "default": 1},
                 "end_line": {"type": "integer", "description": "Last line (inclusive)"},
                 "max_lines": {"type": "integer", "description": "Line cap if no end_line", "default": 500},
@@ -1772,21 +1781,26 @@ LOCAL_TOOLS_DEFINITIONS = [
         "inputSchema": {
             "type": "object",
             "properties": {
-                "path": {"type": "string", "description": "Path to the file to write"},
-                "content": {"type": "string", "description": "Full content of the file"}
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "Path to the file to write"},
+                "content": {"type": "string", "maxLength": MAX_HTTP_BODY_BYTES, "description": "Full content of the file"}
             },
             "required": ["path", "content"]
         }
     },
     {
         "name": "edit_file",
-        "description": "Replace old_str with new_str in an existing file.",
+        "description": (
+            "Replace old_str with new_str in an existing file. "
+            "For several changes, in this file or in others, emit several edit_file calls in the same response; "
+            "they are applied in order, each one on the result of the previous one."
+        ),
         "inputSchema": {
             "type": "object",
             "properties": {
-                "path": {"type": "string", "description": "File path"},
-                "old_str": {"type": "string", "description": "Exact text to replace; must occur once"},
-                "new_str": {"type": "string", "description": "Replacement text"}
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "File path"},
+                "old_str": {"type": "string", "maxLength": MAX_HTTP_BODY_BYTES, "description": "Exact text to replace; must occur once unless replace_all is true"},
+                "new_str": {"type": "string", "maxLength": MAX_HTTP_BODY_BYTES, "description": "Replacement text"},
+                "replace_all": {"type": "boolean", "description": "Replace every occurrence of old_str (e.g. to rename an identifier)", "default": False}
             },
             "required": ["path", "old_str", "new_str"]
         }
@@ -1797,9 +1811,9 @@ LOCAL_TOOLS_DEFINITIONS = [
         "inputSchema": {
             "type": "object",
             "properties": {
-                "query": {"type": "string", "description": "Plain text string or regular expression to search for"},
-                "path": {"type": "string", "description": "Base search directory (defaults to '.')", "default": "."},
-                "file_pattern": {"type": "string", "description": "Optional glob filter (e.g. *.js, *.py)"}
+                "query": {"type": "string", "maxLength": 4096, "description": "Plain text string or regular expression to search for"},
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "Base search directory (defaults to '.')", "default": "."},
+                "file_pattern": {"type": "string", "maxLength": 256, "description": "Optional glob filter (e.g. *.js, *.py)"}
             },
             "required": ["query"]
         }
@@ -1825,6 +1839,7 @@ LOCAL_TOOLS_DEFINITIONS = [
             "properties": {
                 "command": {
                     "type": "string",
+                    "maxLength": MAX_COMMAND_LENGTH,
                     "description": (
                         "PowerShell command to run (e.g. git status; npm test)"
                         if DETECTED_OS == "windows"
@@ -1856,7 +1871,7 @@ LOCAL_TOOLS_DEFINITIONS = [
         "inputSchema": {
             "type": "object",
             "properties": {
-                "path": {"type": "string", "description": "Path to the file or directory to inspect (defaults to '.')"}
+                "path": {"type": "string", "maxLength": MAX_PATH_LENGTH, "description": "Path to the file or directory to inspect (defaults to '.')"}
             }
         }
     },
@@ -1869,11 +1884,12 @@ LOCAL_TOOLS_DEFINITIONS = [
                 "action": {
                     "type": "string",
                     "enum": list(BROWSER_ACTIONS),
+                    "maxLength": 32,
                     "description": "Action to perform in the browser"
                 },
-                "url": {"type": "string", "description": "Target URL for navigate or screenshot"},
-                "selector": {"type": "string", "description": "CSS selector for click or fill actions"},
-                "value": {"type": "string", "description": "Text value to type for fill action"}
+                "url": {"type": "string", "maxLength": 4096, "description": "Target URL for navigate or screenshot"},
+                "selector": {"type": "string", "maxLength": 1024, "description": "CSS selector for click or fill actions"},
+                "value": {"type": "string", "maxLength": MAX_HTTP_BODY_BYTES, "description": "Text value to type for fill action"}
             },
             "required": ["action"]
         }
@@ -1895,11 +1911,8 @@ LOCAL_TOOL_HANDLERS = {
 # Servidor HTTP JSON-RPC 2.0 y SSE con Autenticación por Token
 # ==============================================================================
 
-MAX_HTTP_BODY_BYTES = 1024 * 1024
 MAX_RPC_METHOD_LENGTH = 128
 MAX_TOOL_NAME_LENGTH = 128
-MAX_PATH_LENGTH = 4096
-MAX_COMMAND_LENGTH = 16384
 # Un mensaje MCP admite imágenes en base64, pero no puede agotar la memoria del backend.
 MAX_MCP_MESSAGE_CHARS = 16 * 1024 * 1024
 MAX_MCP_STDERR_LINE_CHARS = 64 * 1024
@@ -1907,73 +1920,32 @@ DEFAULT_MCP_CALL_TIMEOUT_SECONDS = 30
 
 
 def validate_local_tool_arguments(tool_name: str, arguments: dict) -> str | None:
-    """Valida tipos y límites de las herramientas locales antes de ejecutarlas."""
-    schemas = {
-        "list_directory": {
-            "path": (str, MAX_PATH_LENGTH),
-            "recursive": (bool, None)
-        },
-        "read_file": {
-            "path": (str, MAX_PATH_LENGTH),
-            "start_line": (int, None),
-            "end_line": (int, None),
-            "max_lines": (int, None),
-            "max_bytes": (int, None)
-        },
-        "write_file": {
-            "path": (str, MAX_PATH_LENGTH),
-            "content": (str, MAX_HTTP_BODY_BYTES)
-        },
-        "edit_file": {
-            "path": (str, MAX_PATH_LENGTH),
-            "old_str": (str, MAX_HTTP_BODY_BYTES),
-            "new_str": (str, MAX_HTTP_BODY_BYTES)
-        },
-        "run_command": {
-            "command": (str, MAX_COMMAND_LENGTH),
-            "timeout_seconds": (int, None),
-            "background": (bool, None)
-        },
-        "check_command": {
-            "job_id": (int, None),
-            "wait_seconds": (int, None),
-            "stop": (bool, None)
-        },
-        "search_files": {
-            "query": (str, 4096),
-            "path": (str, MAX_PATH_LENGTH),
-            "file_pattern": (str, 256)
-        },
-        "get_diagnostics": {
-            "path": (str, MAX_PATH_LENGTH)
-        },
-        "browser_action": {
-            "action": (str, 32),
-            "url": (str, 4096),
-            "selector": (str, 1024),
-            "value": (str, MAX_HTTP_BODY_BYTES)
-        }
-    }
-    schema = schemas.get(tool_name)
-    if schema is None:
+    """Valida tipos y límites de las herramientas locales a partir de su definición canónica."""
+    definition = next((item for item in LOCAL_TOOLS_DEFINITIONS if item["name"] == tool_name), None)
+    if definition is None:
         return None
 
+    schema = definition.get("inputSchema", {})
+    properties = schema.get("properties", {})
+    type_map = {"string": str, "integer": int, "boolean": bool}
+
     for name, value in arguments.items():
-        expected = schema.get(name)
-        if expected is None:
+        prop = properties.get(name)
+        if prop is None:
             return f"Argumento no permitido: {name}"
-        expected_type, max_length = expected
+        expected_type_name = prop.get("type")
+        expected_type = type_map.get(expected_type_name)
         if expected_type is bool:
             if not isinstance(value, bool):
                 return f"Tipo inválido para '{name}'"
         else:
             if isinstance(value, bool) or not isinstance(value, expected_type):
                 return f"Tipo inválido para '{name}'"
+        max_length = prop.get("maxLength")
         if max_length is not None and len(value) > max_length:
             return f"'{name}' excede el tamaño máximo permitido"
 
-    definition = next((item for item in LOCAL_TOOLS_DEFINITIONS if item["name"] == tool_name), {})
-    for name in definition.get("inputSchema", {}).get("required", ()):
+    for name in schema.get("required", ()):
         if name not in arguments:
             return f"Falta el argumento obligatorio: {name}"
 
@@ -2173,7 +2145,7 @@ class StdioMcpClient:
                     continue
                 try:
                     msg = json.loads(line)
-                except Exception:
+                except json.JSONDecodeError:
                     continue
                 req_id = msg.get("id")
                 if req_id is not None:
@@ -2247,7 +2219,7 @@ class StdioMcpClient:
         if self.running():
             try:
                 self._write(json.dumps({"jsonrpc": "2.0", "method": method}) + "\n")
-            except Exception:
+            except (OSError, ValueError):
                 self._alive = False
 
     def _signal_group(self, sig) -> bool:
@@ -2327,7 +2299,7 @@ class McpServiceManager:
                 server["id"] = server_id
                 server["_directory"] = directory
                 servers[server_id] = server
-            except Exception:
+            except (OSError, json.JSONDecodeError):
                 continue
         return servers
 
@@ -2335,7 +2307,7 @@ class McpServiceManager:
         if self.config_file.exists():
             try:
                 return json.loads(self.config_file.read_text(encoding="utf-8"))
-            except Exception:
+            except (OSError, json.JSONDecodeError):
                 return {}
         return {}
 
@@ -2344,8 +2316,8 @@ class McpServiceManager:
             tmp = self.config_file.with_suffix(".tmp")
             tmp.write_text(json.dumps(self.preferences, indent=2), encoding="utf-8")
             tmp.replace(self.config_file)
-        except Exception:
-            pass
+        except OSError as err:
+            log_event(f"[zerochat] Advertencia al guardar preferencias MCP en {self.config_file}: {err}")
 
     def list_servers(self) -> list[dict]:
         self.services = self._load_services()
@@ -2412,7 +2384,7 @@ class McpServiceManager:
                         installation = json.loads(marker.read_text(encoding="utf-8"))
                         if installation.get("package") != package or installation.get("version") != version:
                             needs_install = True
-                    except Exception:
+                    except (OSError, json.JSONDecodeError):
                         needs_install = True
 
                 if needs_install:
@@ -2596,7 +2568,7 @@ class McpServiceManager:
                 for t in client.tools:
                     try:
                         matches = public_tool_name(server_id, t.get("name")) == public_name
-                    except Exception:
+                    except (ValueError, TypeError):
                         continue
                     if matches:
                         target = (client, t["name"])
@@ -2957,7 +2929,7 @@ class ZeroChatServerHandler(BaseHTTPRequestHandler):
             if content_len <= MAX_HTTP_BODY_BYTES * 2:
                 try:
                     self.rfile.read(content_len)
-                except Exception:
+                except (OSError, ValueError):
                     pass
             self.close_connection = True
             self._reject(413, {"error": "Request body too large"}, safe_path, t0, "Cuerpo HTTP excede el límite")
@@ -3090,7 +3062,7 @@ class ZeroChatServerHandler(BaseHTTPRequestHandler):
                 parsed_out = json.loads(tool_output_json)
                 if isinstance(parsed_out, dict) and parsed_out.get("success") is False:
                     tool_error_info = str(parsed_out.get("error", "Error en herramienta local"))
-            except Exception:
+            except (json.JSONDecodeError, TypeError):
                 pass
             return {"content": [{"type": "text", "text": tool_output_json}], "isError": bool(tool_error_info)}, None, tool_error_info
         if method == "zerochat/project/set_directory":
@@ -3410,6 +3382,8 @@ def main():
     global TOOL_AUTH_KEY, TOOL_AUTH_SESSION_ID, TOOL_AUTH_NONCES
 
     parser = argparse.ArgumentParser(description=f"ZeroChat Local Server v{VERSION}")
+    # Desde Python 3.14 argparse colorea la ayuda con códigos ANSI; la consola la muestra en texto plano.
+    parser.color = False
     parser.add_argument("--port", type=int, default=int(os.environ.get("ZEROCHAT_PORT", DEFAULT_PORT)), help=f"Puerto de escucha (default: {DEFAULT_PORT})")
     parser.add_argument("--host", default=os.environ.get("ZEROCHAT_HOST", DEFAULT_HOST), help=f"Host de escucha (default: {DEFAULT_HOST})")
     parser.add_argument("--ui-url", default=None, help="URL de la interfaz web a abrir (por defecto: interfaz local en desarrollo o GitHub Pages)")