mist-converters 0.1.1

Convert Claude Code session logs (.jsonl) into AgenticGraph JSON

mist-converters — Claude Code Converter

Converts Claude Code session logs (.jsonl files) into AgenticGraph JSON files that capture every LLM call, tool use, thinking step, and subagent invocation as a structured graph.

---

Requirements

• Python 3.9+
• pip

---

Installation

pip install git+https://github.com/AI-Agentic-Tools-Benchmarking/mist-converters.git@claude-convert

This installs the mist-convert CLI command directly — no cloning needed.

---

Running

Convert all projects under ~/.claude/projects (default):

mist-convert

Custom projects directory

Pass the path to your .claude/projects directory:

mist-convert /path/to/your/.claude/projects

Single project directory

mist-convert --dir /path/to/single/project

---

Output

All JSON files are written to claude_code/output/ inside the directory from which you run the script. For example, if you run the script from /home/user/mist-converters, the output will be at /home/user/mist-converters/claude_code/output/. Please send us this full path once the conversion is complete.

claude_code/
└── output/
    └── <project-name>/
        └── <session-id>.json

Each JSON file is an AgenticGraph for one session:

{
  "name": "...",
  "description": "...",
  "stages": [ ... ],
  "edges": [ { "from": "...", "to": "..." }, ... ]
}

---

Project structure

mist-converters/
├── Converter/
│   ├── run_all_projects.sh       ← entry point
│   ├── dir_claude_code_parser.py ← main parser
│   └── tokenizer_utils.py
├── Builder/
│   └── stage.py                  ← AgenticGraph data model
├── token_utils.py                ← tiktoken-based token counter
└── requirements.txt