z/OS FTP MCP Server

A Model Context Protocol (MCP) server for interacting with z/OS mainframe systems via FTP.

Features

• List datasets from mainframe catalog with pattern matching

• Download datasets in binary and text formats with encoding conversion

• Download members from partitioned datasets (PDS) with parallel support

• Submit and monitor JCL jobs

• Advanced text processing: encoding conversion, line ending control, trailing spaces handling

• Configurable write protection for safe production use

• Connection management via environment variables

Installation

For MCP Server Usage

No installation needed! The MCP configuration with uvx will automatically download and run the package.

For Python Library Usage

pip install zos-ftp-mcp

Configuration

Set environment variables for connection:

export ZFTP_HOST="your-mainframe-host"
export ZFTP_PORT="21"
export ZFTP_USER="your-username"
export ZFTP_PASSWORD="your-password"
export ZFTP_TIMEOUT="600.0"
export ZFTP_DOWNLOAD_PATH="/path/to/downloads"
export ZFTP_DEBUG="false"  # Set to "true" for detailed FTP protocol logging
export ZFTP_ALLOW_WRITE="false"  # Set to "true" to enable upload/submit operations

Advanced Text Processing

Control how text files are downloaded from EBCDIC mainframe to ASCII/UTF-8:

# Character encoding conversion (EBCDIC to ASCII/UTF-8)
export ZFTP_DEFAULT_ENCODING="IBM-037,UTF-8"  # US/Canada EBCDIC to UTF-8
# Other options: IBM-1047,UTF-8 (Latin-1), IBM-285,UTF-8 (UK)

# Line ending format
export ZFTP_DEFAULT_LINE_ENDING="LF"  # Unix/Linux style
# Options: CRLF (Windows), LF (Unix), CR (old Mac), NONE

# Preserve trailing spaces in fixed-length records
export ZFTP_PRESERVE_TRAILING_SPACES="false"  # Strip trailing spaces
# Set to "true" to keep 80-character fixed-length records intact

When to use these settings:

• Encoding: Essential for proper character conversion (JCL, COBOL, logs)

• Line endings: Cross-platform compatibility, version control (Git)

• Trailing spaces: Keep "false" for most text files, "true" for fixed-format data

Write Operations

By default, the MCP server is read-only for safety. To enable write operations:

export ZFTP_ALLOW_WRITE="true"

When enabled, you can:

• Upload files to mainframe datasets (upload_dataset)

• Submit JCL jobs (submit_job)

Use with caution in production environments. Read-only operations (list, download, monitor jobs) are always available.

Usage

As MCP Server

{
  "mcpServers": {
    "zos-ftp-mcp": {
      "command": "uvx",
      "args": ["zos-ftp-mcp"],
      "env": {
        "ZFTP_HOST": "your-mainframe-host",
        "ZFTP_USER": "your-username",
        "ZFTP_PASSWORD": "your-password",
        "ZFTP_DOWNLOAD_PATH": "/path/to/downloads"
      }
    }
  }
}

Direct Command Line

zos-ftp-mcp

Tools Available

Dataset Operations

• list_catalog(pattern, limit, offset) - List datasets matching pattern with pagination

• download_binary(source_dataset, target_file) - Download dataset in binary mode

• download_text(source_dataset, target_file, encoding, line_ending, preserve_trailing_spaces) - Download with text processing

• download_pds_members(dataset, target_dir, members, encoding, line_ending, ...) - Download PDS members with advanced options

• upload_dataset(source_file, target_dataset, binary, lrecl, blksize, recfm, space) - Upload file to dataset (requires ZFTP_ALLOW_WRITE=true)

• get_vsam_info(dataset, limit, offset) - Get VSAM dataset information with pagination (requires ZFTP_ALLOW_WRITE=true)

• get_gdg_info(gdg_base, limit, offset) - Get GDG base attributes and generations with pagination (requires ZFTP_ALLOW_WRITE=true)

Job Operations

• submit_job(jcl, jobname) - Submit JCL job (requires ZFTP_ALLOW_WRITE=true)

• list_jes_jobs(jobmask, owner, status, limit, offset) - List jobs with filtering and pagination

• get_job_info(jobid) - Get job details

• download_job_spool(jobid, target_file) - Download job output with return code extraction

Connection Management

• get_connection_info() - Show current connection settings

Sample Usage Prompts

Once configured, you can use these prompts:

Dataset Discovery

"List all datasets starting with SYS1 to explore system datasets"

"Show me all user datasets matching MYUSER.* pattern"

Data Download

"Download the dataset MYUSER.COBOL.SOURCE to my local downloads folder with UTF-8 encoding"

"Download all members from the PDS MYUSER.COBOL.COPYLIB to a local directory"

Data Upload (requires ZFTP_ALLOW_WRITE=true)

"Upload my local file test.jcl to dataset MYUSER.JCL.TEST with lrecl=80 and recfm=FB"

"Upload local file member.cbl to PDS member MYUSER.COBOL.SOURCE(TESTPGM)"

Job Operations

"Submit this JCL job and wait for completion" (requires ZFTP_ALLOW_WRITE=true)

"List all my jobs that are in OUTPUT status"

"Download the spool output for job JOB12345 and show me the return code"

Connection Management

"Show me the current FTP connection settings"

Dependencies

• mcp - Model Context Protocol

• Python 3.10+ standard library (ftplib, pathlib, etc.)

Advanced Features

Pagination Support

To avoid overwhelming context windows with large result sets, several tools support pagination:

Tools with pagination:

• list_catalog - Paginate through datasets

• list_jes_jobs - Paginate through jobs

• get_vsam_info - Paginate through VSAM datasets

• get_gdg_info - Paginate through GDG generations

Usage:

# Get first 50 datasets
list_catalog('USER.*', limit=50)

# Get next 50 datasets
list_catalog('USER.*', limit=50, offset=50)

# Get first 20 jobs
list_jes_jobs(status='OUTPUT', limit=20)

# Get first 10 GDG generations
get_gdg_info('AWS.M2.CARDDEMO.TRANSACT.BKUP', limit=10)

Response includes:

• count - Number of items returned in this page

• total - Total number of items available

• offset - Current offset

• has_more - True if more items are available

JESINTERFACELEVEL Limitations

The server automatically detects your mainframe's JES interface level. With JESINTERFACELEVEL=1 (common in older systems):

Job Submission:

• Jobname in JCL must be userid + exactly one character

• Example: If userid is USER123, jobname must be USER123J

• The server validates this automatically and provides clear error messages

Job Listing:

• Server-side filtering may not work

• Client-side filtering is applied automatically

• Jobs may be purged quickly from spool

Workaround: The server handles these limitations transparently. Just be aware that job retrieval may fail if jobs are purged quickly.

Parallel PDS Downloads

Download PDS members faster using multiple FTP connections:

# Sequential (default)
download_pds_members(dataset, target_dir, members, ftp_threads=1)

# Parallel (2-16 threads)
download_pds_members(dataset, target_dir, members, ftp_threads=4)

Parallel downloads are significantly faster for large PDSs (>10 members).

Per-Operation Overrides

While defaults are set via environment variables, you can override them per operation:

# Use different line ending for specific file
download_text(dataset, target_file, line_ending='CRLF')

# Preserve trailing spaces for fixed-format data
download_text(dataset, target_file, preserve_trailing_spaces=True)

# Use different encoding
download_pds_members(dataset, target_dir, members, encoding='IBM-1047,UTF-8')

License

MIT License - see LICENSE file for details.