Foggy Data MCP Bridge
A semantic layer query engine with MCP support, enabling AI assistants to query structured data through natural language and declarative interfaces.
README
Foggy Data MCP Bridge — Python
A semantic layer query engine with Model Context Protocol (MCP) support, enabling AI assistants to query structured data through natural, declarative interfaces.
Ported from foggy-data-mcp-bridge (Java), maintaining full API compatibility.
What It Does
Foggy Data MCP Bridge sits between your database and AI assistants (Claude, GPT, etc.), providing:
- Semantic Layer — Define business-friendly models (dimensions, measures, calculated fields) on top of raw SQL tables. AI queries "sales by region" instead of writing complex JOINs.
- MCP Protocol — Exposes data through the Model Context Protocol, so AI assistants can discover and query your data models natively.
- Multi-Database — Works with MySQL, PostgreSQL, and SQLite through async drivers.
- Embeddable — Can be vendored into host applications (e.g., Odoo) as a lightweight in-process engine, no separate server required.
┌─────────────────┐ MCP / REST ┌───────────────────────┐
│ AI Assistant │ ◄──────────────────► │ Foggy MCP Bridge │
│ (Claude, etc.) │ JSON-RPC 2.0 │ ┌─────────────────┐ │
└─────────────────┘ │ │ Semantic Layer │ │
│ │ TM/QM Models │ │
┌─────────────────┐ Async SQL │ └────────┬────────┘ │
│ Database │ ◄──────────────────► │ │ │
│ MySQL/PG/SQLite │ aiomysql/asyncpg │ SQL Query Engine │
└─────────────────┘ └───────────────────────┘
Quick Start
Installation
# Clone the repository
git clone https://github.com/foggy-projects/foggy-data-mcp-bridge-python.git
cd foggy-data-mcp-bridge-python
# Install with development dependencies
pip install -e ".[dev]"
# Install database driver(s) you need
pip install aiomysql # MySQL
pip install asyncpg # PostgreSQL
pip install aiosqlite # SQLite (included by default)
Run the Demo Server
# Start with in-memory SQLite demo data
python -m foggy.demo.run_demo --port 8066
# Or connect to an existing database
python -m foggy.mcp.launcher.app \
--db-host localhost --db-port 5432 \
--db-user foggy --db-password secret \
--db-name mydb
Then open http://localhost:8066/docs for the Swagger UI.
Run Tests
python -m pytest --tb=short -q
# 1322 passed, 76 skipped
Architecture
Module Dependency Graph
foggy.mcp (MCP Server, FastAPI)
│
├──► foggy.dataset_model (Semantic Query Engine)
│ ├──► foggy.dataset (SQL Generation, DB Execution)
│ │ └──► foggy.core (Utilities, Exceptions)
│ └──► foggy.fsscript (Expression Engine)
│
└──► foggy.mcp_spi (SPI Types — shared interface layer)
Each layer has strict dependency boundaries — no circular imports, no upward dependencies.
Project Structure
src/foggy/
├── core/ # Utilities, exceptions, filters
├── bean_copy/ # Bean/Map conversion utilities
├── mcp_spi/ # SPI types (shared between all layers)
│ ├── semantic.py # SemanticQueryRequest/Response (Java-aligned Pydantic models)
│ ├── accessor.py # DatasetAccessor, LocalDatasetAccessor
│ ├── enums.py # QueryMode, MetadataFormat, AccessMode
│ └── tool.py # McpTool, ToolResult
├── dataset/ # Database abstraction layer
│ ├── dialects/ # MySQL, PostgreSQL, SQLite, SQL Server
│ ├── db/ # Async executor, connection management
│ └── resultset/ # Record, RecordList
├── dataset_model/ # Semantic layer engine
│ ├── semantic/ # SemanticQueryService (core query engine)
│ ├── engine/ # SQL query builder, formula engine, JOIN graph
│ ├── definitions/ # Model definitions (TM/QM)
│ └── impl/ # Model implementations
├── fsscript/ # FSScript expression engine (ported from Java)
├── mcp/ # MCP server (FastAPI)
│ ├── launcher/ # Application factory, server startup
│ ├── routers/ # HTTP routes (admin, analyst, mcp_rpc, semantic_v3)
│ ├── schemas/ # MCP tool definitions (JSON schema + Markdown)
│ ├── tools/ # Tool implementations (query, metadata, chart)
│ ├── config/ # DataSource, Properties
│ └── audit/ # Tool audit logging
└── demo/ # Demo models and startup scripts
API Reference
MCP Protocol (for AI Assistants)
| Endpoint | Method | Description |
|---|---|---|
/mcp/analyst/rpc |
POST | MCP Streamable HTTP (JSON-RPC 2.0) |
/mcp/analyst/rpc |
GET | SSE stream for server-sent events |
REST API (for Applications)
| Endpoint | Method | Description |
|---|---|---|
/api/v1/models |
GET | List all available models |
/api/v1/models/{name} |
GET | Get model metadata |
/api/v1/query/{name} |
POST | Execute a query |
/api/v1/query/{name}/validate |
POST | Validate query without executing |
/health |
GET | Health check |
/docs |
GET | Swagger UI |
MCP Tools
| Tool | Description | Status |
|---|---|---|
dataset.get_metadata |
Get V3 metadata for all models and fields | ✅ |
dataset.describe_model_internal |
Get detailed metadata for a specific model | ✅ |
dataset.query_model |
Execute a semantic query (V3 payload format) | ✅ |
dataset_nl.query |
Natural language query | ⏳ |
dataset.compose_query |
FSScript multi-model orchestration | ⏳ |
Usage Examples
Query via REST API
# List models
curl http://localhost:8066/api/v1/models
# Query a model
curl -X POST http://localhost:8066/api/v1/query/sales_model \
-H "Content-Type: application/json" \
-d '{
"columns": ["product_name", "total_amount"],
"slice": [{"field": "status", "op": "eq", "value": "confirmed"}],
"groupBy": ["product_name"],
"orderBy": [{"field": "total_amount", "direction": "DESC"}],
"limit": 50
}'
Embedded Mode (No Server)
from foggy.mcp_spi import LocalDatasetAccessor
from foggy.dataset_model.semantic import SemanticQueryService
# Initialize the engine
service = SemanticQueryService(executor=my_db_executor, dialect=my_dialect)
service.register_model(my_table_model)
# Create accessor (accepts standard JSON dict)
accessor = LocalDatasetAccessor(service)
# Query with plain dict — no need to construct typed objects
result = accessor.query_model("sales_model", {
"columns": ["product_name", "total_amount"],
"groupBy": ["product_name"],
"orderBy": [{"field": "total_amount", "direction": "DESC"}],
"limit": 10,
})
# Result is a Pydantic model, serialize to Java-compatible JSON
print(result.model_dump(by_alias=True, exclude_none=True))
MCP Tool Call (JSON-RPC 2.0)
curl -X POST http://localhost:8066/mcp/analyst/rpc \
-H "Content-Type: application/json" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "dataset.query_model",
"arguments": {
"model": "sales_model",
"payload": {
"columns": ["product_name", "total_amount"],
"limit": 10
}
}
}
}'
Database Support
| Database | Driver | Dialect | Status |
|---|---|---|---|
| MySQL | aiomysql | MysqlDialect |
✅ Full support |
| PostgreSQL | asyncpg | PostgresDialect |
✅ Full support |
| SQLite | aiosqlite | SqliteDialect |
✅ Full support |
| SQL Server | — | SqlServerDialect |
🔧 Dialect only |
All database operations are fully async. SQL identifier quoting is dialect-aware (backticks for MySQL, double-quotes for PostgreSQL/SQLite, brackets for SQL Server).
Syncing with Java
Tool definitions (JSON schema + Markdown descriptions) are shared between Java and Python implementations:
# Sync tool definitions from Java project
python scripts/sync_mcp_schemas.py
# Preview changes without applying
python scripts/sync_mcp_schemas.py --dry-run
# Show diff only
python scripts/sync_mcp_schemas.py --diff
Development
Prerequisites
- Python 3.11+
- A database (or use the built-in SQLite demo)
Code Standards
- Type annotations on all public APIs
- Pydantic v2 for data models with Java-aligned camelCase aliases
- Async I/O for all database operations
- No
eval()— SQL parameters use placeholders, never string concatenation - pytest tests required for all new features
Running Checks
# Tests
python -m pytest --tb=short -q
# Type checking
mypy src/foggy/
# Linting
ruff check src/ tests/
Vendoring (Embedded Use)
For embedding in host applications (e.g., Odoo), vendor the minimal module set:
lib/foggy/
├── core/ ✅ Required
├── mcp_spi/ ✅ Required (SPI types + Accessor)
├── dataset/ ✅ Required (SQL engine)
├── dataset_model/ ✅ Required (Semantic engine)
├── fsscript/ ✅ Required (Expression engine)
├── bean_copy/ ✅ Required (Utilities)
└── mcp/ ❌ Not needed (MCP Server, only for standalone deployment)
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.