abl-mcp-server

MCP (Model Context Protocol) server for OpenEdge ABL — pluggable tool architecture with per-project YAML configuration. Provides AI assistants with 24 tools to parse, analyze, lint, document, and scaffold ABL projects.

Built on:

• @breakit/abl-mcp-core — ABL parsing, analysis, linting

• @breakit/abl-mcp-generators — Code scaffolding with REST annotations, ProDataSets, OpenEdge class hierarchy

• @breakit/abl-mcp-contracts — Data contract generators (.i, JSON Schema, TypeScript)

• @breakit/abl-mcp-doc — ABLDoc parsing, doc comment generation, HTML documentation

Quick Start

npx github:breakit/abl-mcp-server

Or add to opencode config:

"abl": {
  "type": "local",
  "command": ["npx", "-y", "github:breakit/abl-mcp-server"],
  "enabled": true
}

Related MCP server: CodeGraphMCPServer

Pluggable Architecture

Each tool is a separate module in src/tools/. Tools are auto-discovered at startup and can be enabled/disabled via a per-project YAML config file.

Per-Project Config (./abl-mcp-server.yaml)

Place this in your ABL project root:

tools:
  enabled:
    - read-abl-file
    - query-abl-symbols
    - analyze-dependencies
    - gen-doc-comment
    - gen-abldoc
    - gen-ablunit-test
    - abl-lint
    # ... add any tools you need
  disabled:
    - check-project-config

All tools are enabled by default. Add names to disabled to turn them off, or set enabled to a specific subset.

Adding Custom Tools

Drop a .ts file into ~/.config/abl-mcp-server/tools/:

import type { ToolModule } from '@breakit/abl-mcp-server/types'

export default {
  name: 'my-custom-tool',
  description: 'Does something custom',
  inputSchema: { type: 'object', properties: { input: { type: 'string' } }, required: ['input'] },
  handler: async ({ input }) => {
    return { content: [{ type: 'text', text: `Got: ${input}` }] }
  },
} satisfies ToolModule

Add my-custom-tool to your abl-mcp-server.yaml enabled list.

Tools (24 total, 13 enabled by default)

All 24 tools are available but only a curated subset is enabled by default. Enable additional tools via abl-mcp-server.yaml (see Pluggable Architecture above).

Default-enabled

Analytical

Generative

Available (disabled by default)

Lint Rules

All 37 rules are defined in config.yaml (shipped with the server). They are inspired by Prolint. To see all active rules: call abl-lint with listRules: true.

Customize rules via your project's abl-mcp-server.yaml:

lint:
  rules:
    # Override severity of an existing rule
    no-undo:
      pattern: '^DEFINE (?:VARIABLE|VAR) +\w+ (?:AS \w+ )?(?!.*NO-UNDO)'
      message: 'DEFINE VARIABLE should include NO-UNDO'
      severity: warning

    # Add a custom rule
    my-naming-convention:
      pattern: '^\s*PROCEDURE\s+[a-z\d]'
      message: 'Procedure names should start with uppercase'
      severity: warning

    # Disable a rule by not including it in enabled (see tools.disabled pattern)

Test Generator Skill

The gen-ablunit-test tool generates skeleton ABLUnit test files. For fully filled test files with realistic assertions and data, an opencode skill is available:

• Skill: llm-fill-ablunit-test

• Workflow: parse source class → generate skeleton → LLM-fills test data + assertions → write completed file

• Requires: opencode with the llm-fill-ablunit-test skill enabled

To install, add to your .opencode/skills/ directory and reference it in your opencode config.

Architecture

abl-mcp-server
├── config.yaml                # Default tool enable/disable + 37 lint rules
├── src/
│   ├── index.ts               # Bootstrap: auto-discovers tools, registers MCP handlers
│   ├── config-loader.ts       # Load + parse per-project YAML config (+ project overlay)
│   ├── types.ts               # ToolModule interface + config types
│   └── tools/                 # 24 pluggable tool modules (auto-discovered)
│       ├── read-abl-file.ts
│       ├── analyze-dependencies.ts
│       ├── abl-lint.ts
│       ├── find-annotations.ts
│       ├── gen-business-entity.ts
│       ├── gen-workflow.ts
│       ├── gen-business-task.ts
│       ├── gen-doc-comment.ts
│       ├── gen-abldoc.ts
│       ├── gen-contract-*.ts
│       └── ...
├── @breakit/abl-mcp-core      # Pure analysis layer — parsers, analysis, linting
├── @breakit/abl-mcp-generators # Scaffolding templates — BE, Service, Controller, Workflow
├── @breakit/abl-mcp-contracts  # Data contract generators — .i, JSON Schema, TypeScript
└── @breakit/abl-mcp-doc        # Documentation utilities — ABLDoc parser + comment generator

Installation

The server is distributed as a GitHub package (not published to npm). Install directly from the repo:

# npm
npm install github:breakit/abl-mcp-server

# pnpm
pnpm add github:breakit/abl-mcp-server

# yarn
yarn add github:breakit/abl-mcp-server

One-shot usage (no install)

npx github:breakit/abl-mcp-server

As an MCP server dependency

Add to your project's package.json:

"dependencies": {
  "@breakit/abl-mcp-server": "github:breakit/abl-mcp-server"
}

Then import in your MCP host:

import { createServer } from '@breakit/abl-mcp-server'

Development

git clone https://github.com/breakit/abl-mcp-server.git
cd abl-mcp-server
yarn install
yarn build
yarn start

Local Multi-Repo Development

If you are working on the sibling repos in ../abl-mcp-core, ../abl-mcp-contracts, ../abl-mcp-doc, and ../abl-mcp-generators, bootstrap them with Yarn and symlink them into this repo:

yarn setup:local

That command:

• runs yarn install in each sibling repo

• builds each sibling repo so their dist/ entrypoints exist

• symlinks them into this repo's node_modules/@breakit/

• includes @breakit/abl-mcp-doc in the local sibling package graph

• links ../abl-mcp-core into ../abl-mcp-generators/node_modules/@breakit/ for local runtime resolution

When you change sibling repo code, rerun:

yarn build:local-deps
yarn link:local-deps

Acknowledgments

• Lint rules inspired by Prolint by Jurjen Dijkstra and contributors

• ABL parsing via tree-sitter-abl

• Language server concepts from abl-language-server

• Naming conventions derived from Progress ABL community standards

License

MIT