Release 01-25-2026

README.md

@@ -31,6 +31,7 @@ Built on top of LangChain's [`SQLDatabase`](https://docs.langchain.com/oss/pytho
 - **Data Visualization**: Generate charts and graphs from query results using natural language (e.g., "show me a bar chart")
 - **Configurable Agents**: YAML-based configuration for adding new data sources
 - **A2A Protocol**: Agent-to-Agent interoperability for integration with other A2A-compliant systems
+- **MCP Protocol**: Model Context Protocol support for Claude Desktop, VS Code, and other MCP clients
 
 ## Architecture
 
@@ -51,7 +52,9 @@ Generates, validates, and executes SQL queries with retry logic.
 - [Database Setup](docs/DATABASE_SETUP.md)
 - [Configuration](docs/CONFIGURATION.md)
 - [Data Visualization](docs/VISUALIZATION.md)
+- [Prompts & Dialects](docs/PROMPTS.md)
 - [A2A Protocol](docs/A2A.md)
+- [MCP Protocol](docs/MCP.md)
 
 ## Quick Start
 

docs/CONFIGURATION.md

@@ -1,19 +1,6 @@
 # Configuration
 
-Data agents are configured via YAML files. See `src/data_agent/config/contoso.yaml` for a complete example.
-
-## Intent Detection
-
-```yaml
-intent_detection_agent:
-  llm:
-    model: gpt-4o
-    provider: azure_openai
-    temperature: 0.0
-  system_prompt: |
-    You are an intent detection assistant...
-    {agent_descriptions}
-```
+Data agents are configured via YAML files. See `src/data_agent/agents/contoso.yaml` for a complete example.
 
 ## Data Agent Definition
 

docs/MCP.md

@@ -0,0 +1,147 @@
+# MCP Protocol Support
+
+The Data Agent supports the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/), enabling integration with Claude Desktop, VS Code, Cursor, and other MCP-compatible clients.
+
+## Quick Start
+
+```bash
+# Start MCP server with SSE transport (default)
+uv run data-agent mcp
+
+# Start with a specific config
+uv run data-agent mcp --config contoso
+
+# Start with stdio transport (for Claude Desktop)
+uv run data-agent mcp --transport stdio
+
+# Start on a custom port
+uv run data-agent mcp --port 9000
+```
+
+## Server Options
+
+| Option | Default | Description |
+|--------|---------|-------------|
+| `--config, -c` | all | Configuration name (e.g., `contoso`). Loads all configs if not specified. |
+| `--transport, -t` | sse | Transport: `sse` for HTTP clients (VS Code, Cursor), `stdio` for Claude Desktop |
+| `--port, -p` | 8002 | Port for SSE transport |
+| `--log-level` | warning | Logging level |
+
+## Available Tools
+
+The MCP server exposes the following tools:
+
+| Tool | Description |
+|------|-------------|
+| `query` | Execute natural language queries against datasources |
+| `list_datasources` | List all configured datasources with descriptions |
+| `list_tables` | List tables for a specific datasource |
+| `get_schema` | Get database schema for a specific datasource |
+| `validate_sql` | Validate SQL syntax without executing |
+
+## Available Resources
+
+| Resource URI | Description |
+|--------------|-------------|
+| `datasources://list` | List of available datasources |
+| `schema://{datasource}` | Database schema for a datasource |
+| `tables://{datasource}` | List of tables for a datasource |
+
+## Client Configuration
+
+### Claude Desktop
+
+Add to `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows):
+
+```json
+{
+  "mcpServers": {
+    "data-agent": {
+      "command": "uv",
+      "args": ["run", "data-agent-mcp"],
+      "cwd": "/path/to/langchain_data_agent"
+    }
+  }
+}
+```
+
+### VS Code
+
+Add to `.vscode/mcp.json` in your workspace:
+
+```json
+{
+  "servers": {
+    "data-agent": {
+      "type": "sse",
+      "url": "http://127.0.0.1:8002/sse"
+    }
+  }
+}
+```
+
+> **Note:** Start the MCP server first with `uv run data-agent mcp` before connecting.
+
+Or for stdio transport (runs server automatically):
+
+```json
+{
+  "servers": {
+    "data-agent": {
+      "type": "stdio",
+      "command": "uv",
+      "args": ["run", "data-agent-mcp", "--transport", "stdio"]
+    }
+  }
+}
+```
+
+### Cursor / Windsurf
+
+Similar configuration to VS Code. Check your IDE's MCP documentation.
+
+## Example Usage
+
+Once configured, you can interact with the Data Agent directly from your AI client:
+
+```
+User: What datasources are available?
+AI: [calls list_datasources] → Shows contoso, adventure_works, amex
+
+User: What's the schema for the contoso database?
+AI: [calls get_schema("contoso")] → Shows tables, columns, types
+
+User: Show me the top 5 products by sales in Q4 2024
+AI: [calls query("top 5 products by sales Q4 2024")] → Returns results
+```
+
+## Programmatic Client Example
+
+```python
+from mcp import ClientSession, StdioServerParameters
+from mcp.client.stdio import stdio_client
+
+async def main():
+    server_params = StdioServerParameters(
+        command="uv",
+        args=["run", "data-agent-mcp"]
+    )
+
+    async with stdio_client(server_params) as (read, write):
+        async with ClientSession(read, write) as session:
+            await session.initialize()
+
+            # List available tools
+            tools = await session.list_tools()
+            print("Available tools:", [t.name for t in tools.tools])
+
+            # Execute a query
+            result = await session.call_tool(
+                "query",
+                arguments={"question": "What are the top selling products?"}
+            )
+            print(result.content)
+
+import asyncio
+asyncio.run(main())
+```

docs/PROMPTS.md

@@ -0,0 +1,173 @@
+# Prompts Module
+
+This module manages system prompts for the Data Agent's LLM interactions. It provides a modular, extensible prompt architecture that supports multiple database dialects and customization through configuration.
+
+## Architecture
+
+```
+src/data_agent/prompts/
+├── __init__.py      # Public exports
+├── builder.py       # Prompt assembly logic
+├── defaults.py      # Default prompt templates
+└── dialects.py      # Database-specific SQL guidelines
+```
+
+## Components
+
+### defaults.py - Core Prompt Templates
+
+Contains the default system prompts used across the agent:
+
+| Prompt | Purpose |
+|--------|---------|
+| `DEFAULT_INTENT_DETECTION_PROMPT` | Routes user questions to the appropriate data agent |
+| `DEFAULT_GENERAL_CHAT_PROMPT` | Handles greetings and capability questions |
+| `DEFAULT_SQL_PROMPT` | Guides SQL generation with schema context |
+| `DEFAULT_RESPONSE_PROMPT` | Formats query results into natural language |
+| `VISUALIZATION_SYSTEM_PROMPT` | Generates matplotlib visualization code |
+| `COSMOS_PROMPT_ADDENDUM` | Cosmos DB-specific constraints and best practices |
+
+### dialects.py - Database-Specific Guidelines
+
+Provides SQL dialect guidelines that are automatically appended based on datasource type:
+
+| Dialect | Datasource Types |
+|---------|------------------|
+| BigQuery | `bigquery` |
+| PostgreSQL | `postgres`, `postgresql` |
+| Azure SQL / SQL Server | `azure_sql`, `mssql`, `sqlserver` |
+| Azure Synapse | `synapse` |
+| Databricks | `databricks` |
+| Cosmos DB | `cosmos`, `cosmosdb` |
+
+Each dialect includes:
+- Syntax conventions (date functions, data types, quoting)
+- Aggregation function usage
+- String manipulation functions
+- Performance best practices
+
+### builder.py - Prompt Assembly
+
+The `build_prompt()` function assembles the final system prompt:
+
+```
+┌─────────────────────────────────────┐
+│     Date Context (current date)     │
+├─────────────────────────────────────┤
+│  Base Prompt (custom or default)    │
+│  - Schema context                   │
+│  - Few-shot examples                │
+├─────────────────────────────────────┤
+│     Dialect Guidelines              │
+│  (based on datasource type)         │
+├─────────────────────────────────────┤
+│  Cosmos Addendum (if applicable)    │
+│  - Partition key constraints        │
+└─────────────────────────────────────┘
+```
+
+## Usage
+
+### Basic Prompt Building
+
+```python
+from data_agent.prompts import build_prompt
+
+# Build a prompt for PostgreSQL
+prompt = build_prompt(
+    datasource_type="postgres",
+    schema_context="Tables: customers (id, name, email), orders (id, customer_id, total)",
+    few_shot_examples="Q: How many customers?\nA: SELECT COUNT(*) FROM customers",
+)
+```
+
+### Custom Prompts via Configuration
+
+Teams can override default prompts in their agent YAML configuration using `system_prompt` and `response_prompt`:
+
+```yaml
+data_agents:
+  - name: my_agent
+    description: E-commerce sales database
+    datasource:
+      type: postgres
+      # ...
+    system_prompt: |
+      You are a SQL expert for our e-commerce database.
+      Focus on sales metrics and customer behavior.
+
+      {schema_context}
+
+      {few_shot_examples}
+    response_prompt: |
+      Provide insights focused on business impact.
+      Always mention revenue implications.
+    table_schemas:
+      # ...
+```
+
+### Getting Dialect Guidelines
+
+```python
+from data_agent.prompts import get_dialect_guidelines
+
+# Get BigQuery-specific SQL guidelines
+guidelines = get_dialect_guidelines("bigquery")
+```
+
+## Prompt Template Variables
+
+The following variables are automatically substituted:
+
+| Variable | Description | Used In |
+|----------|-------------|---------|
+| `{schema_context}` | Database schema information | SQL prompt |
+| `{few_shot_examples}` | Example Q&A pairs | SQL prompt |
+| `{agent_descriptions}` | Available data agents | Intent detection, general chat |
+| `{partition_key}` | Cosmos DB partition key | Cosmos addendum |
+
+## Extending
+
+### Adding a New Dialect
+
+1. Add guidelines constant to `dialects.py`:
+
+```python
+MY_DATABASE_GUIDELINES = """## My Database SQL Guidelines
+
+1. **Syntax conventions:**
+   - Use MY_DATE_FUNC() for date operations
+   - ...
+"""
+```
+
+2. Register in `DIALECT_GUIDELINES_MAP`:
+
+```python
+DIALECT_GUIDELINES_MAP: dict[str, str] = {
+    # ... existing entries
+    "mydatabase": MY_DATABASE_GUIDELINES,
+}
+```
+
+### Adding a New Prompt Type
+
+1. Add the template to `defaults.py`:
+
+```python
+MY_NEW_PROMPT = """You are a specialized assistant for...
+
+{custom_variable}
+"""
+```
+
+2. Export in `__init__.py`:
+
+```python
+from data_agent.prompts.defaults import MY_NEW_PROMPT
+
+__all__ = [
+    # ... existing exports
+    "MY_NEW_PROMPT",
+]
+```

pyproject.toml

@@ -47,12 +47,14 @@ dependencies = [
     "langchain-azure-dynamic-sessions>=0.2.0",
     "matplotlib>=3.10.8",
     "tabulate>=0.9.0",
+    "mcp>=1.25.0",
 ]
 
 [project.scripts]
 data-agent = "data_agent.cli:main"
 data-agent-ui = "data_agent.ui:main"
 data-agent-a2a = "data_agent.a2a.server:main"
+data-agent-mcp = "data_agent.mcp.server:main"
 
 [project.optional-dependencies]
 dev = [