Skip to main content
Glama
khuongtt

codeparse-mcp

by khuongtt

codeparse-mcp

Java/Xtend Code Parser → IR → Graph DB → MCP Server Knowledge base for AI-driven ISO 26262 ASIL-D unit test generation with 100% MC/DC + C0 + C1 coverage.


Architecture (v3)

Java/Xtend Sources
       │
       ├── JavaParser AST Extractor (Java CLI, production)
       │     └── Decision IR (JSON, camelCase)
       │                              │
       ├── Xtend AST Extractor (line-based POC)
       │     └── Decision IR (JSON, camelCase)
       │                              │
       ├── Fallback JS Java Parser (java-parser npm CST)
       │     └── Decision IR (JSON, camelCase)
       │                              │
       └── Fallback JS Xtend Parser (pattern-based)
             └── Decision IR (JSON, camelCase)
                                      │
                                      ▼
                            ┌─────────────────────┐
                            │  ir-ingest.js        │
                            │  Validate IR schema  │
                            │  Compute MC/DC       │
                            │  (centralized, not   │
                            │   duplicated per     │
                            │   parser)            │
                            └──────────┬──────────┘
                                       │
                                       ▼
                            ┌─────────────────────┐
                            │    SQLite Graph DB   │
                            │   (better-sqlite3)   │
                            └──────────┬──────────┘
                                       │
                                       ▼
                            ┌─────────────────────┐
                            │     MCP Server       │
                            │  20 tools via stdio  │
                            │  Compatible with:    │
                            │  • GitHub Copilot    │
                            │  • Claude Desktop    │
                            │  • Any MCP client    │
                            └─────────────────────┘

Key design: AI reads MCP only — never reads raw IR or source files directly. IR is an internal transport format.


Related MCP server: NOMIK

Quick Start

Local

git clone <repo>
cd codeparse-mcp
npm install

# Initialize DB for your project
node src/cli/index.js init --root /path/to/your/project

# Parse all Java/Xtend files
node src/cli/index.js sync --root /path/to/your/project

# Check status
node src/cli/index.js status

Docker

docker build -t codeparse-mcp:latest .

# Init DB
docker run --rm \
  -v /your/project:/project:ro \
  -v codeparse-data:/data \
  codeparse-mcp:latest init

# Sync
docker run --rm \
  -v /your/project:/project:ro \
  -v codeparse-data:/data \
  codeparse-mcp:latest sync

# Status
docker run --rm \
  -v /your/project:/project:ro \
  -v codeparse-data:/data \
  codeparse-mcp:latest status

# Or with docker compose
PROJECT_ROOT=/your/project docker compose run --rm codeparse sync

CLI Commands

Command

Description

codeparse init [--force]

Initialize/reset graph database

codeparse sync [--force]

Parse all files, sync changes only

codeparse status [--errors]

Show graph stats and health

codeparse sync-file <path>

Sync a single file (incremental)

codeparse import-results --junit <path> [--jacoco <path>]

Import JUnit/JaCoCo results

codeparse evidence --asil <level> --class <qname> --output <dir>

Generate ISO 26262 evidence package

Sync Options

codeparse sync \
  --force                          # Re-parse all files
  --include "**/*.java,**/*.xtend" # Custom patterns
  --exclude "**/generated/**"      # Extra exclusions
  --verbose                        # Per-file progress

Import Test Results

# Import JUnit XML + JaCoCo coverage
codeparse import-results \
  --junit build/test-results/test  # JUnit Surefire XML dir
  --jacoco build/reports/jacoco/test/jacoco.xml

Generate Evidence Package

codeparse evidence \
  --asil D \
  --class com.example.SafetyController \
  --output evidence/

Generates 10-file ISO 26262 evidence package:

  • Decision list, MC/DC matrix, test mapping, traceability matrix

  • Coverage gap analysis, audit summary, test specification

  • Requirements cross-reference, review checklist, compliance report


MCP Integration

GitHub Copilot (.vscode/mcp.json)

{
  "servers": {
    "codeparse": {
      "type": "stdio",
      "command": "node",
      "args": ["${workspaceFolder}/src/mcp/server.js"]
    }
  }
}

Claude Desktop (claude_desktop_config.json)

See config/claude-desktop-mcp.json for full example.


MCP Tools (20 total)

Lifecycle

Tool

Description

codeparse_init

Initialize/reset DB

codeparse_sync

Parse and sync all files

codeparse_status

DB health and statistics

sync_file

Sync a single file

Class Queries

Tool

Description

get_class

Full class info (fields, hierarchy, annotations, ASIL)

search_classes

Find classes by name pattern

Method Queries

Tool

Description

get_methods

All methods for a class

search_methods

Find methods by name/signature

get_method_context

Full source context: code, fields, calls, decisions, parse quality

Control Flow Graph

Tool

Description

get_cfg

CFG nodes + edges for a method (C0/C1 coverage)

get_decisions

All decisions with decomposed atomic conditions (decision-UID scoped)

MC/DC (ISO 26262 ASIL-D)

Tool

Description

get_mcdc

MC/DC conditions, truth tables, independence pairs per method

get_mcdc_for_class

All MC/DC data for entire class

Call Graph

Tool

Description

get_callees

Methods called by a method (mock targets)

get_callers

Methods that call a method (impact analysis)

UT Generation (primary AI workflow)

Tool

Description

get_ut_context

Primary tool — full context: class + methods + CFG + MC/DC + field accesses + boundary hints + mock targets in one call

get_dependencies

Import dependencies for a file

Evidence & Coverage (v2.5)

Tool

Description

import_test_results

Import JUnit XML + JaCoCo XML coverage data

get_coverage_summary

Query line/branch/instruction coverage per class/method

export_evidence_plan

Generate 10-file ISO 26262 evidence package


What Gets Parsed

Java

  • Package and import declarations

  • Class/interface/enum/annotation declarations (nested included)

  • All modifiers, annotations, Javadoc

  • Method signatures, parameters, return types, throws

  • Field declarations with types and visibility

  • CFG: if/for/while/do/switch/try/catch/return/throw/break/continue nodes and edges

  • MC/DC: boolean condition decomposition via AST tree parser, truth tables, independence pairs

  • else if kind detection (not plain if)

  • Call graph: method invocations with line numbers

  • Field access tracking (reads and writes via this. / obj.)

  • ASIL level detection from @ASIL_D annotations or /** @ASIL D */ Javadoc

  • Exception types on CATCH/THROW CFG nodes

  • Ternary expressions (cond ? a : b) as decisions

Xtend

  • Package, import, class declarations

  • def, override, dispatch methods

  • val/var fields + Java-style fields

  • All visibility modifiers

  • CFG: if/for/while/do/switch/try/catch/return/throw (pattern-based) with loop body edges

  • MC/DC analysis on boolean expressions

  • Balanced-parenthesis condition extraction (handles nested parens)

  • Extension method detection

  • «IF»/«ELSEIF» template condition parsing

  • Ternary expressions with nested ternary support

  • ASIL annotation detection


Graph DB Schema

The SQLite database stores:

files               → source file registry + SHA-256 for change detection
packages            → Java package index
classes             → class/interface/enum with full metadata + ASIL level
methods             → method signatures + CFG stats + MC/DC summary + ASIL level
fields              → class fields
cfg_nodes           → CFG nodes per method (ENTRY, STATEMENT, BRANCH, LOOP, CATCH, THROW, ...)
cfg_edges           → CFG edges (sequential, true_branch, false_branch, exception, loop_back)
call_edges          → caller → callee relationships
field_accesses      → per-method field read/write tracking (mock/state setup)
dependencies        → file-level import and type dependencies
decisions           → each branch point (if/while/for/ternary/etc.) with expression
conditions          → atomic conditions decomposed from decisions (condition-UID scoped)
mcdc_conditions     → expanded MC/DC with truth tables and independence pairs
mcdc_pairs          → normalized per-condition independence pairs with test vectors
test_cases          → test-to-target method traceability
test_results        → JUnit execution results (from Surefire XML)
coverage_records    → JaCoCo line/branch/instruction coverage
evidence_log        → evidence package generation tracking
parse_errors        → per-file error log

Evidence Export Package (v2.5)

The export_evidence_plan MCP tool (or codeparse evidence CLI) generates 10 ISO 26262 evidence files:

File

Content

01_decision_list.xlsx

All decisions with UIDs, kind, expression, conditions

02_mcdc_matrix.xlsx

MC/DC independence pairs per condition

03_test_mapping.xlsx

Test cases mapped to decisions/conditions

04_traceability_matrix.xlsx

Requirements → decisions → tests

05_coverage_gap_analysis.xlsx

Uncovered branches/pairs analysis

06_audit_summary.md

Audit trail with timestamps

07_test_specification.md

Generated test specification

08_requirements_cross_ref.md

ISO 26262 requirement cross-reference

09_review_checklist.md

Peer review checklist

10_compliance_report.md

ASIL-D compliance summary


Incremental Sync

Files tracked by SHA-256 hash. On each sync:

  • New files → parsed and inserted

  • Changed files → deleted from DB (cascade) and re-parsed

  • Unchanged files → skipped (fast)

  • Call graph → second-pass resolution of caller→callee IDs

  • Content hash prevents re-parsing identical files on repeated runs

Safe to run on every save or in CI without performance penalty.


Extractor Chain (Auto-Fallback)

When parsing a file, the system tries in order:

  1. Java AST Extractor (extractors/java/) — Java CLI tool using JavaParser library, produces full IR

  2. Xtend AST Extractor (extractors/xtend/) — line-based POC, template_if/elseif/ternary support

  3. Fallback JS parserjava-parser.js (CST via npm) or xtend-parser.js (pattern-based)

Extractors produce camelCase IR JSON. Fallback runs if Java is not available. MC/DC pair computation is centralized in ir-ingest.js — not duplicated per parser.


Configuration (.codeparse.json)

{
  "projectRoot": "/path/to/project",
  "dbPath": "/path/to/.codeparse/graph.db",
  "include": ["**/*.java", "**/*.xtend"],
  "exclude": [
    "**/node_modules/**",
    "**/build/**",
    "**/target/**",
    "**/.gradle/**",
    "**/generated/**"
  ]
}

Environment variable override (for Docker): CODEPARSE_PROJECT_ROOT, CODEPARSE_DB_PATH.


Extending: Add More Languages

  1. Add parser in src/parser/<lang>-parser.js exporting parse<Lang>(source, path){ packageName, imports, classes }

  2. Add AST extractor in extractors/<lang>/ producing camelCase IR JSON

  3. Add extension to include globs in config

  4. Register parsing branch in src/graph/builder.js syncProject switch

  5. IR ingest in src/graph/ir-ingest.js handles MC/DC automatically


Known Gaps (P3)

  • No lambda/stream support — .filter().map() logic invisible to CFG

  • No switch expressions (Java 17+ -> arrow cases)

  • No dispatch method support in Xtend parser

  • No Xtend extension method resolution

  • No JML/pre-post condition parsing

  • No requirement/safety-goal traceability table in schema


Requirements

  • Node.js ≥ 20

  • Or Docker (no Node required on host)

  • SQLite (bundled via better-sqlite3)

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    B
    maintenance
    Dev intelligence layer that builds a knowledge graph from any codebase and exposes 7 MCP tools for graph-powered reasoning, impact analysis, and preflight safety and governance checks.
    222 PyPI
    32
    Apache 2.0
  • F
    license
    Not graded
    quality
    D
    maintenance
    AI-native code intelligence graph that builds a persistent knowledge graph of your codebase in Neo4j and exposes it to AI assistants via MCP, enabling contextual code analysis, impact analysis, and dependency tracking.
    23
    -
  • A
    license
    Not graded
    quality
    A
    maintenance
    Provides code intelligence for AI coding agents by indexing repositories into a hybrid knowledge graph, enabling agents to query dependencies, impact, and context through 28 MCP tools.
    3
    Apache 2.0
  • A
    license
    Not graded
    quality
    D
    maintenance
    Enables parsing, indexing, and querying source code as structured knowledge, providing code exploration, spec generation, and migration tools via 20 MCP tools.
    MIT