AI integration¶

.dtsx packages are dense and hard to read by hand, which makes them a natural fit for AI-assisted exploration. Two complementary options ship with this project.

MCP server¶

An optional MCP server exposes the parser as tools to any MCP-compatible client (Claude Desktop, Claude Code, Cursor, …).

pip install "pydtsx-parser[mcp]"

Then register the pydtsx-parser-mcp command with your client:

{
  "mcpServers": {
    "pydtsx-parser": {
      "command": "pydtsx-parser-mcp"
    }
  }
}

Tools provided:

Tool	Purpose
`get_package_summary`	High-level overview — best first call
`get_sql_code`	Extract embedded SQL statements
`get_data_lineage`	Control flow edges plus source → destination tracing
`get_data_flows`	Full data flow component detail and column mappings
`parse_dtsx_file`	Full structured JSON for one file
`parse_ssis_directory`	Full structured JSON for a project

Claude Skill¶

skills/pydtsx-parser/SKILL.md is a portable Agent Skill that teaches Claude when and how to parse SSIS files with the CLI and how to interpret the JSON — no running server required. Copy the skills/pydtsx-parser/ folder into your .claude/skills/ directory to use it.

Which one?¶

Use the MCP server for interactive, on-demand parsing inside a client that cannot run shell commands. Use the Skill for a portable, dependency-light way to give any shell-capable agent the know-how — it pairs with the CLI and jq and needs no server process.

Either way, give the model the LLM context guide when it needs to interpret parser output in depth.