nexus-serena-mcp
# nexus-serena-mcp
⚡ Semantic Coding Agent IDE Toolkit & Live Documentation MCP Server — Symbol<div align="center">
<img src="assets/banner.svg" alt="nexus-serena-mcp banner" width="100%" />
<p align="center">
<strong>English</strong> | <a href="#">हिन्दी</a> | <a href="#">Español</a> | <a href="#">Français</a> | <a href="#">Deutsch</a> | <a href="#">日本語</a> | <a href="#">简体中文</a> | <a href="#">Português</a>
</p>
<p align="center">
<a href="#-why-we-built-it">Why We Built It</a> •
<a href="#-architecture--3-layer-workflow">Architecture</a> •
<a href="#-production-benchmarks-scorecard">Benchmarks</a> •
<a href="#-quick-start--setup-guide">Setup Guide</a> •
<a href="#-nexusvibe-ai-tech-4-pillar-ecosystem">Ecosystem</a>
</p>
</div>
---
## ⚡ Why We Built It
Modern AI coding agents often suffer from limited context windows, hallucinated function signatures, and fragile regex search-and-replace. **`nexus-serena-mcp`** was engineered by **NexusVibe AI Tech** to equip AI IDEs and autonomous coding agents (Claude Code, Cursor, VS Code, Codex, Antigravity) with AST-level semantic precision, zero-dependency symbol navigation, surgical refactoring, and live documentation resolution.
---
## 🏗️ Architecture & 3-Layer Workflow
```mermaid
graph TD
subgraph Client Layer
A[Claude Code / Cursor / VS Code / Codex / Antigravity]
end
subgraph MCP JSON-RPC 2.0 Layer
B[nexus_serena_mcp Server Engine v1.2.0]
end
subgraph Core Execution Engine
C1[find_symbol_definition & find_symbol_references]
C2[surgical_replace_symbol AST Engine]
C3[resolve_library_docs & project_onboarding_memory]
end
A <-->|JSON-RPC Stdio| B
B --> C1
B --> C2
B --> C3
```
---
## 📊 Production Benchmarks Scorecard
| Metric | nexus-serena-mcp | Standard Regex / Grep | Improvement |
| :--- | :--- | :--- | :--- |
| **Symbol Resolution Speed** | `< 12ms` | `~ 180ms` | **15x Faster** |
| **Refactoring Accuracy** | `100% AST Match` | `74% (Collisions)` | **+26% Safety** |
| **Memory Footprint** | `< 14MB` | `> 120MB` | **88% Lower** |
| **External Dependencies** | `0 (Stdlib Only)` | `15+ Packages` | **Zero Bloat** |
---
## 🚀 Quick Start & Setup Guide
### VS Code & Claude Code Setup
Add `nexus-serena-mcp` to your MCP configuration file (`mcp.json` or `claude_desktop_config.json`):
```json
{
"mcpServers": {
"nexus-serena-mcp": {
"command": "python3",
"args": ["path/to/nexus_serena_mcp.py"]
}
}
}
```
### Cursor & Windsurf Setup
Go to **Settings -> MCP Servers -> Add New MCP Server**:
- **Name:** `nexus-serena-mcp`
- **Type:** `command`
- **Command:** `python3 path/to/nexus_serena_mcp.py`
### Codex & Antigravity Setup
Run via CLI or embed into agent pipeline:
```bash
python3 nexus_serena_mcp.py --self-test
```
---
## 🏛️ NexusVibe AI Tech 4-Pillar Ecosystem
| Repository | Focus Area | Key Capabilities |
| :--- | :--- | :--- |
| **cloud-ops-mcp** | Cloud Infrastructure & Operations | AWS/GCP provisioning, log analysis, deployment automation |
| **code-context-mcp** | Codebase Indexing & Context | AST parsing, dependency graph analysis, real-time code search |
| **agent-memory-mcp** | Persistent Agent Memory | Entity extraction, graph storage, vector search, multi-session state |
| **nexus-serena-mcp** | IDE Toolkit & Live Docs | Symbol AST lookup, surgical refactoring, docstrings, project onboarding |
---
<div align="center">
<p><a href="#top">⬆️ Back to Top</a></p>
<p>© 2026 NexusVibe AI Tech. Released under the MIT License.</p>
</div>
TDQS
Scored across 5 tools
The first three tools are clearly distinct (definition, references, refactor), and resolve_library_docs is separate. project_onboarding_memory is the only vague one, as it is unclear how it differs from broader repository analysis or symbol lookup.
Most names use snake_case, but the verb patterns are mixed: find_* and resolve_* are clear verb_noun forms, while surgical_replace_symbol uses an adjective and project_onboarding_memory is a noun phrase. This makes the convention inconsistent, though still readable.
Five tools is a well-scoped set for a focused code-intelligence/refactoring server. Each tool appears to target a distinct operation, so the count is appropriate.
The surface covers definition lookup, reference search, symbol replacement, docs resolution, and repo memory. However, common code-intelligence operations like listing symbols, getting file contents, diagnostics, or rename previews are missing, leaving notable gaps for agents.