mainframe-ingest-mcp 0.1.1

MCP server for ingesting COBOL mainframe codebases — classifies files, resolves dependencies, maps CICS transactions to REST routes

mainframe-ingest-mcp

An MCP server that ingests COBOL mainframe codebases from ZIP files and produces a structured JSON manifest — ready for downstream modernization tools.

pip install mainframe-ingest-mcp

---

What is this?

Mainframe modernization projects always start with the same problem: nobody knows what's in the codebase. Thousands of COBOL files, copybooks, BMS screens, JCL jobs, and CSD exports — all sitting in a ZIP someone exported from the mainframe.

mainframe-ingest-mcp solves the discovery step. Point it at any mainframe ZIP and it will:

• Classify every file by type (COBOL, copybook, BMS screen, JCL, DB2 DDL, CSD export)
• Detect and convert EBCDIC encoding to UTF-8
• Scan every COPY, EXEC SQL INCLUDE, EXEC CICS, and CALL statement
• Build a full dependency graph (which programs use which copybooks)
• Parse CICS transaction definitions and suggest FastAPI REST routes
• Map every BMS screen to an Angular component name
• Produce a single manifest.json — the handoff document for the rest of the pipeline

It has been tested against a real 294-file, 118,000-line government CICS application and resolves 99%+ of dependencies with zero configuration.

---

Installation

pip install mainframe-ingest-mcp

Requires Python 3.10+.

---

Connecting to Claude Desktop

This is an MCP server — it runs locally and exposes tools to Claude Desktop.

Step 1: Find your Claude Desktop config file:

OS	Location
macOS	~/Library/Application Support/Claude/claude_desktop_config.json
Windows	%APPDATA%\Claude\claude_desktop_config.json
Linux	~/.config/Claude/claude_desktop_config.json

Step 2: Add this to the config:

{
  "mcpServers": {
    "mainframe-ingest-mcp": {
      "command": "mainframe-ingest-mcp"
    }
  }
}

Step 3: Restart Claude Desktop.

That's it. The tools will appear automatically. You can now say to Claude:

"Extract the ZIP at /Users/me/PAYROLL_SYSTEM.zip to /tmp/payroll and build a manifest"

Claude will call the tools in the right order and return the full manifest.

---

Usage (without Claude Desktop)

You can also call the tools directly from Python:

from zip_ingestion_mcp.tools.manifest import build_file_manifest
from zip_ingestion_mcp.tools.extractor import extract_zip

# Extract the ZIP
extract_zip("/path/to/PAYROLL_SYSTEM.zip", "/tmp/payroll")

# Build the manifest
manifest = build_file_manifest("/tmp/payroll", "/path/to/PAYROLL_SYSTEM.zip")

import json
print(json.dumps(manifest["summary"], indent=2))

Output:

{
  "total_files": 294,
  "cobol_programs": 83,
  "copybooks": 177,
  "bms_screens": 17,
  "cics_transactions": 304,
  "total_cobol_lines": 77962
}

---

Available MCP Tools

Tool	What it does
tool_build_file_manifest	The main tool. Runs everything and returns the full manifest.
tool_extract_zip	Extracts a ZIP file safely (zip-slip protected)
tool_detect_all_artifacts	Classifies every file by type
tool_detect_encoding	Detects if a file is EBCDIC or UTF-8
tool_convert_encoding	Converts EBCDIC → UTF-8 (creates .bak backup)
tool_scan_dependencies	Builds a full COPY/CICS/SQL dependency graph
tool_parse_csd_export	Parses CICS transaction definitions → REST route suggestions

In most cases you only need tool_build_file_manifest — it calls all the others internally.

---

The manifest.json

The manifest is a structured JSON document designed to be consumed by downstream modernization tools:

{
  "manifest_version": "1.0",
  "readiness": "READY",
  "summary": {
    "total_files": 9,
    "cobol_programs": 2,
    "copybooks": 4,
    "bms_screens": 1,
    "cics_transactions": 3,
    "total_cobol_lines": 156,
    "ebcdic_files_detected": 0,
    "unresolved_copybooks": 0
  },
  "cobol_programs": [
    {
      "file": "cbl/online/EMPINQRY.cob",
      "line_count": 89,
      "dependencies": {
        "copies": ["EMPRECRD", "DFHAID"],
        "bms_sends": ["EMPINQMS"],
        "sql_includes": []
      }
    }
  ],
  "bms_screens": [
    {
      "mapset_name": "EMPINQMS",
      "file": "bms/EMPINQMS.bms",
      "angular_component": {
        "component_class": "EmpinqmsComponent",
        "selector": "app-empinqms",
        "template_file": "empinqms.component.html"
      }
    }
  ],
  "route_table": [
    {
      "transaction_code": "EMPI",
      "program": "EMPINQRY",
      "http_method": "GET",
      "fastapi_route": "/api/employee/{id}"
    }
  ],
  "warnings": [],
  "next_steps": [
    "Send 2 COBOL programs to cobol-parser-mcp",
    "Send 1 BMS screen to bms-to-angular-mcp"
  ]
}

---

File types supported

Extension	Classified as	Notes
.cbl, .cob, .cobol	COBOL_PROGRAM	Main business logic
.cpy	COPYBOOK or BMS_COPYBOOK	Detected by content
.bms	BMS_SCREEN	3270 terminal screen
.jcl	JCL_JOB	Batch job definitions
.ddl, .sql	DB2_DDL	DB2 schema definitions
.csd	CSD_EXPORT	CICS transaction registry
.inc, .include	SQL_INCLUDE or JCL_INCLUDE	Detected by content

---

Bundled IBM system copybooks

The following IBM system copybooks are bundled with the package because they live in IBM system libraries on the mainframe and are never included in application ZIPs:

• DFHAID — CICS attention identifier keys (PF1-PF24, ENTER, CLEAR, PA1-PA3)
• SQLCA — SQL Communication Area (SQLCODE, SQLERRM, SQLWARN)

These are resolved automatically — no configuration required.

---

Part of a larger pipeline

mainframe-ingest-mcp is Step 1 of a COBOL → modern stack modernization pipeline:

ZIP file
   └── mainframe-ingest-mcp      ← you are here
         └── manifest.json
               ├── cobol-parser-mcp       → extracts business logic
               ├── bms-to-angular-mcp     → generates Angular components
               └── db2-schema-mcp         → converts DB2 → PostgreSQL

---

License

MIT