RPG代码检查器

ILE-RPG&DSPF Code Checker

An MCP server that checks compliance with ILE-RPG syntax and coding standards.

Overview

This MCP server analyzes ILE-RPG source code and detects issues such as:

• Specification order errors
• Column position rule violations
• Line continuation errors
• Naming convention violations
• Use of deprecated features
• Best practice violations

Features

Check Functions

1. Structure Check

   • Specification order (H→F→D→P→I→C→O) with P-procedure section awareness
   • Column position rules for each specification (H/F/D/P/C)
   • D-spec advanced validation (name continuation ..., declaration type placement, trailing period detection)
   • Line length limit (maximum 100 columns)
   • DBCS (Double-Byte Character Set) support with shift character consideration

2. Syntax Check

   • Line continuation rule validation
   • Prohibition of multiple statements on one line
   • /FREE and /END-FREE correspondence
   • Correct use of **FREE format
   • Parentheses and string literal matching

3. Naming Convention Check

   • Variable and procedure name conventions
   • Special character usage restrictions
   • Meaningful name recommendations
   • Reserved word checking (strict level)

4. Best Practice Check

   • Detection of 16 deprecated features (GOTO, TAG, MOVE, MOVEL, Z-ADD, Z-SUB, etc.)
   • Warning for numbered indicators (*IN01-*IN99)
   • Recommendation for fully free format
   • Custom rule support

5. Common Error Detection

   • F-spec space shortage
   • D-spec column position errors (including column shift heuristics)
   • Continuation line errors

Check Levels

• basic: Basic checks (specification order, critical syntax errors)
• standard: Standard checks (basic + column positions, line continuation, etc.)
• strict: Strict checks (standard + naming conventions, best practices)

Multi-language Support

• English (en): Default language
• Japanese (ja): Full Japanese support for all messages and reports

DBCS Support

When considerDBCS option is enabled, the checker accounts for shift-in (SI) and shift-out (SO) characters used in DBCS strings (Japanese, Chinese, Korean). This ensures accurate column position validation when DBCS characters are present.

Custom Rules

Users can define custom best practice rules in a JSON file. Rules support:

• Pattern-based matching (regex)
• Custom severity levels
• Enable/disable functionality
• Rule descriptions and suggestions

Installation

Prerequisites

• Node.js 18 or higher
• npm

Setup

1. Install dependencies:

cd /path/to/ilerpg-code-checker
npm install

2. Build:

npm run build

3. Add to MCP configuration file:

For IBM Bob IDE

Add the following to c:\Users\user\AppData\Roaming\Bob-IDE\User\globalStorage\ibm.bob-code\settings\mcp_settings.json:

{
  "mcpServers": {
    "ilerpg-code-checker": {
      "command": "node",
      "args": ["<path-to-project>\\build\\index.js"],
      "disabled": false,
      "alwaysAllow": [],
      "disabledTools": []
    }
  }
}

For Claude Desktop

Add to your Claude Desktop MCP settings file (typically ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "ilerpg-code-checker": {
      "command": "node",
      "args": ["<path-to-project>/build/index.js"]
    }
  }
}

For Cline (VS Code Extension)

Add to your Cline MCP settings in VS Code:

{
  "mcpServers": {
    "ilerpg-code-checker": {
      "command": "node",
      "args": ["<path-to-project>/build/index.js"]
    }
  }
}

Note: Replace <path-to-project> with the actual absolute path to your ilerpg-code-checker directory.

Usage

MCP Tools

Note on Token Efficiency: When checking files, prefer using check_rpg_file over check_rpg_code. The check_rpg_file tool reads files directly on the server side, significantly reducing token usage compared to passing entire code content through check_rpg_code. This is especially important for large files or when checking multiple files.

1. check_rpg_code

Checks entire RPG source code. Note: For better token efficiency, consider using check_rpg_file when checking files.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_rpg_code",
  arguments: {
    code: "H DFTACTGRP(*NO)\n...",
    checkLevel: "standard",  // "basic" | "standard" | "strict"
    language: "en",          // "en" | "ja"
    considerDBCS: false,     // true to consider DBCS shift characters
    customRulesPath: "./custom-rules.json"  // Optional
  }
})

Return value:

{
  "valid": false,
  "issues": [
    {
      "severity": "error",
      "category": "structure",
      "line": 10,
      "column": 7,
      "message": "Invalid specification order...",
      "rule": "SPEC_ORDER",
      "suggestion": "..."
    }
  ],
  "summary": {
    "totalIssues": 5,
    "errors": 2,
    "warnings": 2,
    "infos": 1,
    "checkedLines": 100
  }
}

2. check_specification_order

Checks specification order only.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_specification_order",
  arguments: {
    code: "H DFTACTGRP(*NO)\n...",
    language: "en"  // "en" | "ja"
  }
})

3. check_column_positions

Checks column position rules.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_column_positions",
  arguments: {
    code: "H DFTACTGRP(*NO)\n...",
    language: "en",       // "en" | "ja"
    considerDBCS: false   // true to consider DBCS shift characters
  }
})

4. check_naming_conventions

Checks naming conventions.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_naming_conventions",
  arguments: {
    code: "D MyVar           S             10A\n...",
    language: "en"  // "en" | "ja"
  }
})

5. check_best_practices

Checks best practices.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_best_practices",
  arguments: {
    code: "C                   GOTO      Label\n...",
    language: "en",                         // "en" | "ja"
    customRulesPath: "./custom-rules.json"  // Optional
  }
})

6. check_rpg_file

Checks file by file. Recommended: This tool is more efficient than check_rpg_code as it reduces token usage by reading files directly on the server side.

use_mcp_tool({
  server_name: "ilerpg-code-checker",
  tool_name: "check_rpg_file",
  arguments: {
    filePath: "//your-server/path/to/your-file.rpg",
    checkLevel: "standard",
    language: "en",          // "en" | "ja"
    considerDBCS: false,     // true to consider DBCS shift characters
    customRulesPath: "./custom-rules.json"  // Optional
  }
})

Custom Rules

Create a JSON file to define custom best practice rules:

{
  "rules": [
    {
      "id": "no-select-all",
      "name": "Avoid SELECT *",
      "description": "Using SELECT * is not recommended",
      "pattern": "SELECT\\s+\\*",
      "severity": "warning",
      "suggestion": "Specify column names explicitly",
      "enabled": true
    },
    {
      "id": "require-error-handling",
      "name": "Error handling required",
      "description": "All file operations should have error handling",
      "pattern": "CHAIN|READ|WRITE|UPDATE|DELETE",
      "severity": "info",
      "suggestion": "Add error handling with %ERROR or *IN99",
      "enabled": true
    }
  ]
}

Custom Rule Properties

• id: Unique identifier for the rule
• name: Display name
• description: Detailed description
• pattern: Regular expression pattern to match
• severity: "error" | "warning" | "info"
• suggestion: Recommendation for fixing the issue
• enabled: true | false

Examples of Detected Issues

Structure Errors

F-spec
H-spec  ← Error: H-spec must be placed before F-spec

Column Position Errors

D MyVariable    S             10A  ← Error: Variable name is not in correct column position

Line Continuation Errors

D MyLongVariableName
D-                    S             10A  ← Error: Continuation line spec type mismatch

Naming Convention Violations

D x               S             10A  ← Warning: Variable name is too short
D 1stVar          S             10A  ← Error: Variable name cannot start with a number

Deprecated Features

C                   GOTO      Label  ← Warning: GOTO is deprecated
C     Label        TAG

DBCS Example

When considerDBCS: true:

D MyVar           S             10A   INZ('日本語')
  ↑ Correctly accounts for shift characters in column position check

The checker also validates EBCDIC byte length once DBCS shift (SO/SI) bytes are counted:

• Comment lines (* fixed-form or // free-form) over 80 EBCDIC bytes → warning (COMMENT_LINE_EBCDIC_OVERFLOW). Past 80 bytes the compile listing garbles; past 100 bytes the source is truncated by CPYFRMSTMF (reported as LINE_LENGTH error).
• Comment area (columns 81-100) over 20 EBCDIC bytes → warning (COMMENT_DBCS_OVERFLOW).

Development

Project Structure

ilerpg-code-checker/
├── package.json
├── tsconfig.json
├── README.md                 # Documentation (English)
├── README.ja.md              # Documentation (Japanese)
├── DESIGN.md                 # Architecture & design document
├── custom-rules.example.json # Custom rules example
├── src/
│   ├── index.ts              # MCP server entry point
│   ├── orchestrator.ts       # Check orchestration
│   ├── parser/
│   │   ├── rpg-parser.ts     # RPG code parser
│   │   └── line-analyzer.ts  # Line analysis
│   ├── checkers/
│   │   ├── structure-checker.ts     # Spec order & column validation
│   │   ├── syntax-checker.ts        # Syntax & format validation
│   │   ├── naming-checker.ts        # Naming convention validation
│   │   ├── best-practice-checker.ts # Deprecated features & best practices
│   │   └── common-errors-checker.ts # Common error patterns
│   ├── i18n/
│   │   └── messages.ts       # Multi-language messages (en/ja)
│   ├── config/
│   │   └── custom-rules.ts   # Custom rules manager
│   ├── types/
│   │   └── index.ts          # Type definitions
│   └── utils/
│       ├── file-reader.ts    # File I/O operations
│       ├── reporter.ts       # Report formatting
│       └── dbcs-helper.ts    # DBCS support utilities
└── build/                    # Compiled output

Build

npm run build

Development Mode (Auto-build)

npm run watch

Troubleshooting

MCP Server Won't Start

1. Check Node.js version: node --version (18 or higher required)
2. Verify build is complete: Check if build/index.js exists
3. Verify MCP configuration file path is correct

Check Results Not Displayed

1. Check source code encoding (UTF-8 recommended)
2. Verify file path is correct
3. Check MCP server logs

License

MIT

Author

guricat

Version

0.0.17

Changelog

0.0.17 (2026-08-07)

• The server now reads its version from package.json at runtime
  • The hardcoded version in src/index.ts ('0.0.15') was missed in the v0.0.16 release, so package.json (0.0.16) and the server's reported version (0.0.15) disagreed (reported by a consuming project). From now on only the package.json version needs updating
• Resolved dependency vulnerabilities (npm audit: 8 → 0)
  • @modelcontextprotocol/sdk 1.25.3 → 1.30.0 (GHSA-345p-7cg4-v4c7: cross-client data leak via shared server/transport instance reuse), plus transitive dependency updates (hono / @hono/node-server / fast-uri / path-to-regexp / ajv / body-parser). All within semver-compatible ranges, so only package-lock.json changed
• No change to checker detection logic. All four regression suites pass; all five samples remain at 0 errors

0.0.16 (2026-06-16)

• Added detection of comment lines exceeding 80 EBCDIC bytes: COMMENT_LINE_EBCDIC_OVERFLOW (warning, only when DBCS handling is enabled)
  • Comment lines with Japanese (DBCS) text reaching 80–100 EBCDIC bytes slipped through every length check (checkLineLength only errors past 100 bytes; checkCommentArea only covers the col81-100 area and skips comment lines. Real case: an 86-byte // comment line)
  • // free-form comments also evaded the isComment flag (col7 ≠ '*'), so comment-line detection was extended to "col7 = '*' or the line starts with //"
  • Past 80 bytes the compile listing garbles; past 100 bytes CPYFRMSTMF truncates the source. Lines over 100 bytes are left to the existing LINE_LENGTH error to avoid double reporting
• Added regression test tests/test-comment-ebcdic.mjs. All five samples remain at 0 errors

0.0.15 (2026-06-16)

• Repository housekeeping (no change to checker detection logic)
  • Fixed the typo in the sample directory name (saples/ → samples/)
  • Consolidated the three test scripts (test-samples / test-dspec-overflow / test-dspf-overlap) under tests/. Added a single-source verification helper tests/check-one.mjs <path>
• Sanitized and compressed the sample sources (~4660 lines → ~1500 lines total)
  • Neutralized to avoid evoking the original subject matter and renamed to SAMPLE01–SAMPLE05. Algorithm bodies stubbed out, identifiers and comments genericized
  • Preserved the diversity of RPG constructs (the value for column-position checks) and kept all five files compile-clean (0 errors)
  • Sanitized proprietary references in the bundled docs as well (generic RPG/EBCDIC knowledge retained)

0.0.14 (2026-06-12)

• Added leading attribute byte collision detection to DSPF_FIELD_OVERLAP (real case: AITNTCFGD.dspf CTLWIN row 4 — '状態:'(4,2)=7 bytes col2-8 and 'テナント'(4,9) are adjacent without data overlap, yet CPD7866 was issued)
  • On 5250 displays, the position immediately preceding each field is occupied by its leading attribute byte, so at least one blank column is required between the end of one field's data and the start of the next. A gap of 0 (starting at end+1) places the attribute byte on the previous field's data and causes CPD7866. The check was extended from next.col <= current.endCol to next.col <= current.endCol + 1, with a dedicated message for attribute-byte collisions
  • A gap of 1 (starting at end+2) is valid since the attribute byte fits in the blank column (no false positive)
• Fixed halfwidth katakana misclassification in DBCSHelper: U+FF00-FFEF was treated as DBCS wholesale, doubling the byte length of halfwidth katakana (U+FF61-FF9F, which is SBCS = 1 byte via code page 290 in EBCDIC 5026/5035). Surfaced as a false positive on '日次ﾄｰｸﾝ上限 (0=無制限):' (AIOPSD.dspf) once the attribute-byte check was added. Halfwidth forms (U+FF61-FFDC, U+FFE8-FFEE) are now SBCS
• Added 5 regression test cases (DBCS-constant attribute-byte collision / gap-1 clean / field starting right after an SBCS constant / its fix / halfwidth-katakana constant). Verified 0 false positives across all 10 DSPF files in E:/IBM_i_and_AI/src. During verification, the checker also caught CWOPT(14,12) in the user's in-progress AITNTCFGD CTLWIN edit as a true positive (attribute byte colliding with > at col11 of '選択 ==>')

0.0.13 (2026-06-11)

• Fixed DSPF_FIELD_OVERLAP (CPD7866-equivalent) misses (real case: in AIADMND.dspf TENFTR, the overlap between the unconditioned constant 'F3/12=戻る F5=再表示' and the indicator-77 conditioned constant 'F6=追加' went undetected)
  • Constants (literals) are now included in overlap detection: previously only named fields were checked, so constant-vs-constant and constant-vs-field overlaps (the typical CPD7866 pattern) all passed through. Display length is computed deterministically as the post-EBCDIC byte length (n DBCS characters = 2n+2 bytes including SO/SI). Constant continuation lines ('+' = continue from the first non-blank of the next line, '-' = continue from column 45 of the next line) are supported
  • Indicator exclusion narrowed to "both sides conditioned" only: previously any indicator on either side skipped the check, but CPD7866 means an overlap with a field without conditioning. Only mutually exclusive pairs like 77/N77 (CPD7865-equivalent, intentional) are excluded; an overlap with an unconditioned side is reported
• Added CJK Symbols and Punctuation (U+3000-303F: ideographic space, 「」, 。, ・ etc.) to DBCSHelper as DBCS (byte length was previously underestimated)
• Added regression test test-dspf-overlap.mjs (pre-fix AIADMND overlap detected / post-fix clean / exclusive-indicator pairs not reported / continued-literal length). Verified 0 false positives across all 10 DSPF files in E:/IBM_i_and_AI/src and no change in results for all RPG sources

0.0.12 (2026-06-11)

• Resolved the column-22 E ambiguity using context, closing most of the v0.0.11 blind spot where an overflowing name happened to end with E at column 22
  • PR/PI parameter lines: D lines with a blank definition type following a PR/PI declaration are now tracked as parameters. Since column 22 must be blank in parameter definitions even when it is E (RNF3780), E is reported as name overflow (D_SPEC_NAME_OVERFLOW) or an invalid value (D_SPEC_EXT_DESC_INVALID) with certainty
  • Externally described DS lines: E + no EXTNAME + a name of 11+ characters cannot compile under either interpretation (without EXTNAME the name field is used as the file name, max 10 characters) → new rule D_SPEC_EXTNAME_REQUIRED (error)
  • E on subfield lines outside PR/PI blocks (externally described subfields) remains out of scope, as it is indistinguishable from valid code
• Expanded the regression test to 20 cases (added multi-line PR/PI block context cases); verified 0 false positives across all E:/IBM_i_and_AI/src sources

0.0.11 (2026-06-11)

• Added D-spec name overflow detection: D_SPEC_NAME_OVERFLOW (error)
  • Detects names that overflow the name field (columns 7-21) into column 22 and beyond (e.g. a 14-character name starting at column 9). Previously undetectable when the remaining fields stayed at their correct columns (real case: P_DBC_CONSUMED in AIMONR.sqlrpgle caused RNF3780 severity 20)
  • Provides a Before/After correction using the ... name-continuation syntax
• Added value validation for column 22 (External Description) and column 23 (Type of Data Structure): D_SPEC_EXT_DESC_INVALID / D_SPEC_DS_TYPE_INVALID (error)
  • Per the IBM ILE RPG Reference, column 22 allows only E/blank and column 23 allows only S (PSDS) / U (data-area DS) / blank; non-blank values outside a DS definition cause RNF3780, so these are reported as errors
  • Also detects S/U in column 23 when columns 24-25 are not DS
• D_SPEC_DECL_TYPE_MISPLACED no longer fires when column 21 (the last name-field column) is occupied (reported as name overflow instead)
• Verified 0 false positives across all known-good E:/IBM_i_and_AI/src sources; added regression test test-dspec-overflow.mjs

0.0.10 (2026-06-07)

• Removed the erroneous F_SPEC_SPACING check (false positives on valid code)
  • Column 17 of an F-spec is the File Type field (I/O/U/C), not a required space; a 10-character file name (e.g. AICHATHSTD) is followed directly by the File Type letter, which is valid and compiles
• Downgraded D_SPEC_NAME_POSITION from warning to info
  • The column-7/column-8 start convention is a style recommendation, not a compile requirement; a name placed anywhere within columns 7-21 compiles cleanly
• Verified against the known-good E:/IBM_i_and_AI/src sources (*.sqlrpgle / *.dspf): 0 errors / 0 warnings at standard and strict levels

0.0.9 (2026-06-06)

• Fixed INDICATOR_USAGE (numbered indicator) false positives
  • Excluded F-spec OFLIND(*INxx) (PRINTER overflow indicator), since *INxx is mandatory by language spec and cannot be a named indicator
  • Excluded *INxx strings inside comment lines (col 7 = *, e.g. F*/C*/D*)

0.0.8 (2026-06-06)

• Renamed display name to "ILE-RPG&DSPF Code Checker" (internal identifier unchanged)
• Fixed DSPF/DDS A-spec false positives by aligning column parsing with the IBM DDS layout (name 19-28, length 30-34, line 39-41, position 42-44)
• Scoped field-overlap detection to within each record format (R)
• Excluded A-spec from quote/parenthesis matching (DDS uses different continuation rules)
• Distinguished WINDOW definition vs reference; detect standalone OVERLAY (no parentheses)
• Limited the DBCS position-shift warning to the fixed-column area (cols 1-44)
• Excluded CNTFLD (continued fields) from line-overflow checks

0.0.7 (2026-02-12)

• Fixed false positives for WCA4i test cases
• Improved parser robustness for continuation lines
• Fixed SPEC_ORDER false positives inside P-procedure sections

0.0.6 (2026-02-12)

• Added Phase 9 compile findings for D-spec validation
• Name continuation syntax (...) handling
• Declaration type misplacement detection (col 22-23 vs col 24-25)
• Trailing period detection in field names (RNF0622/RNF0623)
• Column shift heuristics for keyword detection in size field

0.0.5 (2026-02-12)

• Enhanced D-spec column position validation
• More robust column position checking across specifications
• Improved error messages and suggestions

0.0.4 (2026-01-25)

• Fully free-format (**FREE) detection
• Token efficiency improvements in tool descriptions
• Fixed **FREE detection before line length validation

0.0.3 (2026-01-25)

• Added fully free-format detection support
• Tool description improvements for token efficiency
• Renamed to "ILE RPG Code Checker"
• Added version number to startup message

0.0.2 (2026-01-25)

• Updated documentation
• Changed title from "ILE-RPG Coding Standards Checker" to "ILE-RPG Code Checker"
• Updated description to include "syntax and coding standards"
• Masked sensitive information in examples
• Fixed DBCS example to use INZ instead of Japanese variable names

0.0.1 (2026-01-25)

• Initial release
• Basic check functionality implementation
• 6 MCP tools provided
• Multi-language support (English/Japanese)
• DBCS character support
• Custom rules functionality

References

• IBM i Information Center - ILE RPG Reference

Support

For issues or questions, please report to the project's Issue tracker.