# CHANGELOG
All notable changes will be tracked here. Dates reflect when the feature set landed in the repository; see git history for exact commits.
## v0.5.2 – Fixed Gradio MCP Tool Exposure (December 2, 2025)
### Critical Bug Fix
- **Gradio MCP server now properly exposes only 3 tools** (`mt5_query`, `mt5_analyze`, `mt5_execute`) instead of all internal helper functions
- Added `mcp_functions` parameter to explicitly control which functions are exposed as MCP tools
- Previously, Gradio was auto-exposing internal UI helper functions (`get_server_status`, `update_history`, `update_positions`, `_lambda_`, etc.) as MCP tools, cluttering the tool list
### What Changed
- **Before**: 9+ tools exposed (3 actual + 6+ internal UI functions)
- **After**: Exactly 3 tools exposed (mt5_query, mt5_analyze, mt5_execute)
- **Impact**: Clean MCP API surface, no confusion about which tools to use
## v0.5.1 – Production-Ready Error Handling (December 2, 2025)
### Comprehensive Error Handling
- **New Error Utilities Module** (`error_utils.py`):
- Standardized error types (12 categories: JSONParseError, ValidationError, TypeError, ValueError, EnumError, MissingFieldError, MT5ConnectionError, MT5OperationError, CalculationError, TimeoutError, RuntimeError, UnknownError)
- Safe JSON parsing with detailed error messages and position tracking
- Safe enum conversion with valid values listing in error responses
- Field validation helpers for required fields and type checking
- Standardized error response format with timestamp and operation context
- **Tool Function Protection**:
- All MCP tools (mt5_query, mt5_analyze, mt5_execute) wrapped with comprehensive error handling
- Input validation: command length limits (50KB), dangerous operation detection, type checking
- Rate limit error handling with graceful degradation
- MT5 connection validation before all operations
- Catch-all exception handlers to prevent server crashes
- **Handler-Level Protection**:
- Request object validation in `handle_mt5_query` and `handle_mt5_analysis`
- Parameter type validation before MT5 operations
- Custom error re-raising with wrapped exceptions for better debugging
- **Connection Resilience**:
- Retry logic for MT5 initialization (3 attempts with 1-second delay between retries)
- Detailed error logging for connection failures
- Namespace building error handling with fallback messages
- **Executor Safety**:
- Command and namespace validation before execution
- Code preparation error handling with syntax validation
- Result formatting error handling with fallback to type information
- **Security Enhancements**:
- Dangerous operation blocking: eval(), exec(), subprocess, mt5.initialize(), mt5.shutdown()
- Command length limits to prevent abuse (50KB maximum)
- SQL injection attempt detection and handling
- **Test Coverage**: 30+ test cases covering all error scenarios with 100% pass rate
- Malformed JSON inputs handled gracefully
- Invalid enum values return helpful error messages
- Type mismatches caught and reported
- Missing required fields validated
- MT5 connection failures retry and report
### What Changed
- **Before**: Server crashed on malformed client input (invalid JSON, wrong types, missing fields)
- **After**: All errors handled gracefully with standardized JSON error responses
- **Impact**: Server is now production-ready and stable under malformed/malicious inputs
## v0.5.0 – Multi-Transport MCP Server
- Added a Gradio v6-powered MCP server that exposes the same MT5 tools over streamable HTTP/SSE (`/gradio_api/mcp/`) with optional hosting on Hugging Face Spaces or Windows VPS targets.
- Introduced a dual-transport launcher with `--transport stdio|http|both`, customizable host/port/rate-limit flags, and stdio-only default behavior for backward compatibility.
- Implemented per-IP HTTP rate limiting (10 req/min by default) plus shared MT5 connection locking to keep the terminal stable under concurrent requests.
- Documented new workflows in README/USAGE, including MCP client snippets, deployment recipes, and migration guidance from v0.4.x.
- Cleaned up planning artifacts and unused files in preparation for publishing to `main` and PyPI.
## v0.4.0 – Forecasting & ML Signals
- Added Prophet forecasting support inside `mt5_analyze`, including `periods`, `freq`, `seasonality_mode`, `growth`, `plot`, and `plot_components` controls.
- Introduced optional XGBoost-based trading signals (`enable_ml_prediction`, `ml_lookback`, and detailed signal metadata with confidence scores, feature provenance, and training sample counts).
- Extended namespace dependencies (`prophet`, `xgboost`, `scikit-learn`) and refreshed quick-start flows for daily and hourly forecasting scenarios.
- Enhanced forecast responses with structured summaries (trend direction, percentage change, uncertainty ranges) plus automatically saved chart paths for downstream visualization.
## v0.3.0 – Structured Input Tools
- Delivered the `mt5_query` tool for JSON-based MT5 operations with strict validation (symbol existence, timeframe parsing, parameter typing) and LLM-friendly corrective suggestions.
- Released `mt5_analyze`, enabling one-shot pipelines that combine MT5 data retrieval, 80+ `ta` indicators, chart generation (single or multi-panel), and optional Markdown/JSON/chart-only outputs.
- Added systematic error messaging, function signature caching, and parameter normalization to reduce retries from AI assistants and improve throughput.
## v0.2.1 – LLM Compatibility Fixes
- Reworked tool descriptions with highly visible guardrails that explain the pre-initialized MT5 session and the requirement to assign results before returning.
- Updated every bundled example to fetch data, convert to pandas explicitly, and store outputs in `result` to avoid silent failures.
- Added runtime inspection that blocks `mt5.initialize()` / `mt5.shutdown()` (and MetaTrader5 equivalents) before execution and returns actionable guidance.
- Raised expected success rates from ~37% to 87%+ across test models by combining documentation fixes, validation, and clearer error messaging.
## v0.1.0 – Initial Release
- Launched the `execute_mt5` tool with a read-only MetaTrader 5 bridge, safe namespace construction, and automatic formatting of dicts, tables, and named tuples.
- Implemented connection management, optional file logging, and comprehensive multi-line execution support for AI coding assistants.
