🌉 StackBridge-MCP
Sub-1ms Cross-Stack AST Contract & Verification Layer for AI Coding Agents
-brightgreen.svg)
Quick Start •
Client Config •
Benchmarks •
MCP Tools •
CLI Reference •
Docs
💡 Why StackBridge?
When AI coding agents (Cursor, Claude Code, Windsurf, Antigravity) edit backend models or API routes in full-stack codebases, backend unit tests frequently pass while the frontend silently breaks in production:
- An agent modifies an API parameter or Pydantic/SQLAlchemy field in
backend/routes.py. - Backend tests pass in isolation. Nothing warns the agent.
- The React/Next.js client calling that endpoint across the boundary fails with runtime errors.
StackBridge-MCP is an always-warm Model Context Protocol (MCP) server that parses full-stack AST relationships, discovers cross-stack blast radii in 0.75 ms, and verifies changes using baseline-diffed compiler checks with zero false positives.```
React / Next.js Client FastAPI Routes SQLAlchemy ORM Models
(TypeScript AST) ───► (Python AST) ───► (Schema AST)
UserProfile.tsx get_user_billing() BillingAccount
---
## ⚡ Key Highlights
- **🌲 Tree-sitter AST Graph:** Parses Next.js (`fetch`, Axios, React Query) ↔ FastAPI routes ↔ SQLAlchemy ORM models without heavy LSP sidecars or runtime imports.
- **⚡ Sub-1ms Traversal:** Persistent SQLite WAL database with recursive Common Table Expressions (**0.75 ms** traversal query latency).
- **📉 99.74% Prompt Token Reduction:** Replaces massive multi-file code dumps with compact, mathematically precise AST contract slices.
- **🛡️ Root-Cause Diagnostic Ranking:** Graph-distance BFS ranks errors (`🔴 PRIMARY ROOT CAUSE` vs `⚠️ CASCADING BREAKAGE`) and outputs immediate Git diff patches.
- **🧪 Test Impact Selection:** Isolates test suites impacted by a schema change and highlights untested blast-radius paths (0% coverage).
- **🌐 Interactive Canvas:** Built-in localhost tripartite visualizer (`stackbridge ui`) on `http://127.0.0.1:3456`.
- **🔄 Continuous Intelligence:** Background file watcher daemon (`stackbridge watch`) and living `AGENTS.md` context generator.
---
## 📊 Real-World Benchmarks
Empirical performance measured on **`fastapi-realworld-example-app`** (44 files, 23 AST dependency nodes, 10 cross-boundary edges):
| Benchmark Metric | Raw Codebase Dump | StackBridge Compact Slice | Improvement / Latency |
| :--- | :--- | :--- | :--- |
| **Context Window Size** | `19,705 tokens` | `51 tokens` | 📉 **99.74% Token Reduction** |
| **Blast Radius Traversal** | Full-repo search: `~150 ms` | SQLite Recursive CTE: `0.75 ms` | ⚡ **200x Faster Traversal** |
| **Compiler Verification** | Global linter: `~3,500 ms` | Baseline-Diffed Engine: `312 ms` | 🛡️ **Zero False Positives** |
| **Automated Test Suite** | — | 56 / 56 tests passing | ✅ **100% Passing** |
*See full benchmark methodology in `docs/benchmarks.md` and `REAL_WORLD_BENCHMARK.md`.*
---
## 🚀 Quick Start
### Option 1: Zero-Install Execution (Recommended via `uvx`)```bash
uvx stackbridge serve
Option 2: Pip Installation```bash
pip install stackbridge
stackbridge serve
---
## ⚙️ Client Configuration
Connect StackBridge to your AI pair programmer over standard JSON-RPC 2.0 stdio:
### 1. Cursor (`.cursor/mcp.json`)```json
{
"mcpServers": {
"stackbridge": {
"command": "uvx",
"args": ["stackbridge", "serve"]
}
}
}
2. Claude Desktop (claude_desktop_config.json)```json
{
"mcpServers": {
"stackbridge": {
"command": "python",
"args": ["-m", "stackbridge.main", "serve", "--transport", "stdio"]
}
}
}
---
## 🤖 MCP Tools Reference
StackBridge exposes high-ergonomics tools to coding agents:
| Tool Name | Arguments | Description |
| :--- | :--- | :--- |
| `trace_fullstack_path` | `symbol_or_path: str` | Traces the full-stack dependency chain: Frontend component ➔ API route ➔ Database model. |
| `get_route_contract` | `route_path: str` | Extracts HTTP methods, status codes, response models, and linked frontend fetch callers with confidence scores. |
| `verify_schema_change`| `modified_files: dict` | Runs in-memory compiler checks across impacted files, ranking root causes and proposing diff patches. |
| `get_stack_health` | `repo_path: str` | Returns real-time full-stack boundary stats, node counts, edge counts, and breakage drift status. |
---
## 💻 CLI Reference```bash
# Index a repository and export the dependency graph
stackbridge index --repo-path . --force
# Trace blast radius for a model or route
stackbridge trace --target BillingAccount
# Run pre-commit boundary verification guard
stackbridge guard --fail-on-error
# Launch interactive tripartite web visualizer
stackbridge ui --port 3456
# Start continuous background watcher daemon
stackbridge watch
# Generate living AGENTS.md boundary architecture guide
stackbridge init-agents
# Execute performance and token reduction benchmarks
stackbridge benchmark --runs 3 --output BENCHMARK.md
📁 Repository Structure```text
StackBridge-MCP/
├── .github/
│ ├── workflows/ci.yml # CI pipeline (Python 3.10-3.13 on Ubuntu/Windows/macOS)
│ ├── ISSUE_TEMPLATE/ # Bug report and feature request issue templates
│ └── PULL_REQUEST_TEMPLATE.md # Standard PR checklist
├── docs/
│ ├── architecture.md # Subsystem breakdown and Mermaid diagrams
│ ├── benchmarks.md # Benchmark methodology and raw metrics
│ └── ast_extraction_spec.md # Tree-sitter extractor grammar specifications
├── stackbridge/
│ ├── core/ # Unified StackGraph, SQLite CTE store, watcher, route matcher
│ ├── parsers/ # Tree-sitter parsers (TS fetch, Python routes, SQLAlchemy)
│ ├── verifier/ # Baseline-diffed verifier, root-cause ranker, test impact selector
│ ├── mcp_server/ # FastMCP stdio server and JSON-RPC tools
│ ├── benchmarks/ # Benchmark runner and markdown report generator
│ └── ui/ # Localhost tripartite interactive canvas
├── tests/ # 56 automated test suites (parsers, verifiers, MCP E2E, CTE)
├── AGENTS.md # Living agent architecture guide
├── CHANGELOG.md # Version release notes
├── CONTRIBUTING.md # Contribution and development guidelines
├── LICENSE # MIT License
└── pyproject.toml # Package metadata and tool configurations
```
📄 License
This project is licensed under the MIT License.
