xslt-mcp

An MCP server for XSLT — spec reference lookup, stylesheet validation, and XSLT/XPath execution powered by the PhoenixmlDb engine.

Install

Option 1: Self-contained binary (no .NET required)

Download the latest release for your platform from GitHub Releases:

# Linux/macOS
chmod +x xslt-mcp-linux-x64
./xslt-mcp-linux-x64

Option 2: .NET tool (requires .NET 10 SDK)

dotnet tool install -g xslt-mcp

Configure

Add to your MCP client configuration:

Claude Code (.mcp.json):

{
  "mcpServers": {
    "xslt": {
      "command": "xslt-mcp"
    }
  }
}

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "xslt": {
      "command": "xslt-mcp",
      "args": []
    }
  }
}

Tools

Spec Reference

Execution

Utilities

How It Works

The MCP server is built from two things: spec reference files and the PhoenixmlDb XSLT engine.

Spec Files

The spec/ directory contains Markdown files with YAML frontmatter covering XSLT 3.0/4.0, XPath 3.1, and Functions & Operators. Each file documents one instruction, function, error code, or concept:

spec/
  instructions/     # 74 files — xsl:template, xsl:for-each-group, xsl:stream, ...
  xpath/
    functions/      # 160+ files — fn:tokenize, map:merge, math:sqrt, ...
    axes/           # 13 files — child, descendant, following-sibling, ...
  concepts/         # 15 files — streaming-rules, patterns, serialization, ...
  error-codes/      # 23 files — XTSE0010, XTDE0640, FODC0002, ...
  output-methods/   # 7 files — xml, html, json, text, xhtml, csv, adaptive

Every file has this structure:

---
name: xsl:for-each-group
category: instruction
since: "2.0"
spec_url: https://www.w3.org/TR/xslt-30/#for-each-group
---

# xsl:for-each-group

Groups items in a sequence and processes each group.

## Attributes
...

At startup, the server loads all Markdown files, parses the YAML frontmatter, and builds an in-memory index. The lookup tools search this index by name, category, or full-text keywords. The content is returned directly to the AI agent as Markdown.

Execution Engine

The execution tools (xslt_transform, xslt_validate, xpath_evaluate) delegate to the PhoenixmlDb.Xslt engine — a .NET implementation of XSLT 3.0/4.0. This means the MCP server can not only look up how a feature should work, but also run it and show you the actual output.

Embedded Resources

When installed as a dotnet tool or self-contained binary, the spec files are embedded as assembly resources — no external files needed. For local development, you can override the spec path:

XSLT_SPEC_PATH=/path/to/spec xslt-mcp
# or
xslt-mcp --spec-path /path/to/spec

Spec Coverage and Contributing

The spec reference files were AI-generated from the W3C XSLT 3.0, XPath 3.1, and Functions & Operators specifications. They are interpretive summaries derived from the published standards — useful as reference for AI agents, but not a verbatim reproduction of the spec text. Every entry includes a spec_url linking to the authoritative W3C source.

We're actively filling gaps and improving accuracy. The initial release was generated from an LLM's training data without systematically checking against the spec's table of contents, which means coverage skewed toward frequently-discussed features and missed advanced/niche areas. We've since audited against the full spec and filled the major gaps (streaming, merge, accumulators, 100+ functions), but there's always room to improve.

Areas where we'd welcome help

• Streaming rules — The streamability categories (grounded, striding, climbing, etc.) are complex. Our streaming-rules.md covers the fundamentals but the full spec has nuances we may have simplified.
• Output serialization parameters — The interaction between xsl:output, xsl:result-document, and character maps across all output methods has many edge cases.
• Schema-aware processing — Type annotations, xsl:import-schema, and validation interactions are lightly covered.
• Error code completeness — The XSLT 3.0 spec defines 100+ error codes. We have 23. Each one matters when you're debugging.
• XSLT 4.0 additions — xsl:map, xsl:map-entry, and other 4.0 features need coverage as the spec stabilizes.

How to contribute

1. Fork this repo
2. Add or edit files in spec/ — follow the existing YAML frontmatter + Markdown format
3. Submit a PR

The only requirements are: the name, category, and since frontmatter fields must be present, and the content should be accurate to the W3C spec. If you're unsure about a detail, include a note and link to the relevant spec section — an incomplete-but-honest entry is better than a confident-but-wrong one.

License

Apache-2.0