mcp-plesk-dev-docs 0.5.0

A unified MCP server that indexes and retrieves Plesk documentation using vector embeddings and semantic search with reranking

mcp-plesk-dev-docs

State-of-the-Art (SOTA) semantic search across the entire Plesk documentation surface, optimized for sub-second latency on Apple Silicon.

---

Why this exists

Plesk documentation is spread across five separate sources: an admin guide, a REST API reference, a CLI reference, a PHP SDK, and a JS SDK. Answering a single extension development question often means searching all of them manually, cross-referencing results, and still missing the relevant section.

This server ingests all five sources, embeds them with a multilingual model, and exposes a single search_plesk_unified MCP tool. It uses hybrid search (Vector + FTS), Reciprocal Rank Fusion (RRF), and Cross-Encoder reranking to deliver high-precision results in milliseconds.

---

Architecture & Performance

flowchart TD
    Client["MCP Client\n(Claude Desktop / Cursor / etc.)"]

    Client -->|"search_plesk_unified(query)"| Server

    subgraph Server["FastMCP Server · Modular Architecture"]
        direction TB
        Main["Bootstrap · server/main.py"]
        Life["Lifecycle Hooks · server/lifecycle.py"]
        Tools["MCP Tools · server/mcp_app.py"]

        Main --> Life --> Tools
    end

    subgraph Pipeline["Retrieval Pipeline"]
        direction TB
        E["1 · Embed query\n(Hardware-accelerated)"]
        S["2 · Hybrid Search\nVector (LanceDB) + FTS (Tantivy)"]
        R["3 · RRF Merge + Rerank\n(MiniLM-L4-v2)"]
        N["4 · Neighbor Expansion\n(Context Enrichment)"]
        A["5 · AI Synthesis\n(sampling-enabled)"]
        E --> S --> R --> N --> A
    end

    subgraph Store["LanceDB Vector & FTS Store"]
        direction LR
        G["Guide"]
        A_["API"]
        C["CLI"]
        P["PHP Stubs"]
        J["JS SDK"]
    end

    Tools --> Pipeline
    S <--> Store

Performance Benchmarks (2026-05-04)

Optimized for Apple Silicon (M2/M3) using MPS acceleration and memory-resident table caching.

Profile	Embed Model	HR@5	MRR@5	Avg Latency	Est. RAM
light	BAAI/bge-small	100.0%	0.917	1.007 s	~200 MB
medium	BAAI/bge-base	100.0%	0.917	~0.60s	~600 MB
full-tq	BAAI/bge-m3	75.0%	0.750	~0.40s	~1300 MB

Metrics measured on Apple M2 Pro with LanceDB connection caching enabled.

---

Key Features

• Sub-Second Hybrid Search: Combined Vector + Tantivy FTS with RAM-cached table connections for instant retrieval.
• AST-Aware Chunking: Uses tree-sitter to respect class and method boundaries in PHP, JS, and TS documentation.
• TurboQuant Acceleration: Fast 4-bit quantized search for the full-tq profile, delivering 10x lower latency for large models.
• Neighborhood Retrieval: Automatically fetches adjacent chunks (prev/next) to provide complete context for grounding.
• Macro-Context Summaries: Injects file-level purpose summaries into every chunk using the SummaryCache.
• AI-Synthesized Answers: Generates concise answers from search results with structured inline citations [1], [2].

---

MCP Components

This server provides tools, prompts, and resources. See docs/mcp-components.md for a full reference.

Primary Tools

Tool	Description
search_plesk_unified	Hybrid search with RRF and Cross-Encoder reranking.
get_file_content	Retrieve the full content of a specific documentation file.
resolve_references	Find all files referencing a specific symbol or topic.
refresh_knowledge	Re-fetch sources and update the index (incremental).
trigger_index_sync	Start a background indexing job.
daemon_health	Check readiness, hardware acceleration (MPS/CUDA), and latency stats.

Resources

• plesk://toc/api - Table of Contents for API documentation.
• plesk://toc/cli - Table of Contents for CLI reference.
• plesk://toc/guide - Table of Contents for Extensions Guide.
• plesk://toc/php-stubs - Hierarchical list of PHP classes.

---

Quickstart

Install

git clone https://github.com/barateza/mcp-plesk-dev-docs.git
cd mcp-plesk-dev-docs
uv pip install -e .

Initial Indexing

uv run python -m mcp_plesk_dev_docs.server.main refresh_knowledge

Running

# Standard mode
uv run python -m mcp_plesk_dev_docs.server.main

# Responsive daemon mode (auto-warmup)
PLESK_DAEMON_AUTO_WARMUP=true uv run python -m mcp_plesk_dev_docs.server.main

---

Configuration

Set environment variables in .env:

PLESK_MODEL_PROFILE=light       # light | medium | full-tq
PLESK_ENABLE_SAMPLING=true     # AI-Synthesized answers
PLESK_DAEMON_AUTO_WARMUP=true  # Preload models on startup
PLESK_INDEX_SUMMARIES=true     # Enable file-level summaries
OPENROUTER_API_KEY=sk-or-v1-...

---

Documentation

• docs/benchmarks.md - Detailed latency and quality reports.
• docs/mcp-components.md - Full tool and resource reference.
• docs/turboquant.md - 4-bit quantization internals.

---

License

MIT. See LICENSE.

Ownership & Disclaimer

This is a personal project by Gilson Siqueira. It is not officially affiliated with, endorsed by, or supported by Plesk or WebPros International GmbH. Plesk is a trademark of WebPros International GmbH.

Important notice about Plesk-owned deliverables

Portions of this repository were developed under contract for Plesk International GmbH ("Plesk") only if specifically identified as such. The MIT license above applies only to material the repository owner is authorized to license. Files or directories owned by Plesk, if any, are listed in NOTICE. If you need assurance about licensing for a particular file, contact Plesk or seek legal counsel before relying on the MIT License for Plesk-owned files.

Built to make Plesk extension development faster.