session-convert

session-convert

Enables conversion of coding agent sessions between Claude Code, Codex CLI, and OpenCode, including full support for OpenCode's SQLite storage.

Category
访问服务器

README

session-convert

MCP server that converts coding agent sessions between Claude Code, Codex CLI, and OpenCode.

The only tool that supports OpenCode SQLite session conversion — agent-convert and ctxmv handle Claude↔Codex but nothing touches OpenCode's .opencode/opencode.db.

What problem does this solve?

You're using multiple AI coding agents (Claude Code, Codex CLI, OpenCode) and want to migrate a session from one to another — maybe you started in Claude Code but want to continue in OpenCode, or you found a great Codex session that would be useful in Claude.

Each agent stores sessions differently:

  • Claude Code — JSONL files in ~/.claude/projects/<hash>/sessions/
  • Codex CLI — Rollout JSONL in ~/.codex/sessions/YYYY/MM/DD/
  • OpenCode — SQLite database in <project>/.opencode/opencode.db

This MCP server reads from any of these, normalizes into a unified format, and writes to any other. All 6 conversion directions work (Claude→Codex, Claude→OpenCode, Codex→Claude, Codex→OpenCode, OpenCode→Claude, OpenCode→Codex).

Quick Start

git clone https://github.com/megamen32/session-convert.git
cd session-convert
npm install
npm run build

MCP Configuration

Add to your MCP client config (Claude Desktop, Cursor, Windsurf, etc.):

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "session-convert": {
      "command": "node",
      "args": ["/absolute/path/to/session-convert/dist/index.js"]
    }
  }
}

Cursor / Windsurf (.cursor/mcp.json or equivalent):

{
  "mcpServers": {
    "session-convert": {
      "command": "node",
      "args": ["/absolute/path/to/session-convert/dist/index.js"]
    }
  }
}

MCP Tools

list_sessions

List available sessions from a harness. Filter by working directory prefix.

list_sessions({ harness: "claude", cwdPrefix: "~/apps" })
list_sessions({ harness: "codex" })
list_sessions({ harness: "opencode", searchPaths: ["~/projects"] })

convert_session

Convert a session from one harness to another by session ID.

convert_session({ from: "claude", to: "opencode", sessionId: "abc-123" })
convert_session({ from: "codex", to: "claude", sessionId: "xyz-789" })
convert_session({ from: "opencode", to: "codex", sessionId: "oc-session-1", projectPath: "/path/to/project" })

preview_conversion

Dry-run a conversion — shows message breakdown, tool call stats, and compatibility warnings without writing anything.

preview_conversion({ from: "claude", to: "opencode", sessionId: "abc-123" })

convert_by_path

Convert a session file by its direct path. Auto-detects the format (Claude JSONL, Codex rollout JSONL, or OpenCode .db).

convert_by_path({ sourcePath: "/home/user/.claude/projects/.../sessions/abc.jsonl", to: "codex" })
convert_by_path({ sourcePath: "/home/user/myproject/.opencode/opencode.db", to: "claude" })

Conversion Matrix

From \ To Claude Code Codex CLI OpenCode
Claude Code — ✅ ✅
Codex CLI ✅ — ✅
OpenCode ✅ ✅ —

What Gets Preserved

Content type Claude → Codex Claude → OpenCode Codex → Claude Codex → OpenCode OpenCode → Claude OpenCode → Codex
Text messages ✅ ✅ ✅ ✅ ✅ ✅
Tool calls (name + args) ✅ ✅ ✅ ✅ ✅ ✅
Tool results ✅ ✅ ✅ ✅ ✅ ✅
Thinking / reasoning ✅ ✅ ✅ ✅ ✅ ✅
Images ✅ ✅ ✅ ✅ ✅ ✅
Session metadata (CWD, timestamps) ✅ ✅ ✅ ✅ ✅ ✅
Model name ✅ ✅ ✅ ✅ ✅ ✅
Token usage Approx Approx ✅ Approx Approx Approx

How It Works

Source format  →  Unified Conversation  →  Target format
(JSONL/SQLite)     (typed Messages)         (JSONL/SQLite)
  1. Read — Parse the source session (Claude JSONL, Codex rollout JSONL, or OpenCode SQLite) into a unified Conversation object with typed Message parts (text, tool_call, tool_result, thinking, image)
  2. Transform — Map content types between format-specific representations (e.g., Claude's tool_use ↔ Codex's function_call ↔ OpenCode's double-wrapped tool_call part)
  3. Write — Serialize to the target harness's native format and store it in the correct location

Format-Specific Details

  • Claude Code stores sessions as one JSON object per line in .jsonl files. Tool calls are tool_use blocks, results are tool_result blocks in the next user message. Thinking blocks are stored as thinking content type.
  • Codex CLI uses a session_meta header line followed by response_item entries. Tool calls are separate function_call items with matching function_call_output items. Reasoning is a separate reasoning item type.
  • OpenCode uses SQLite with a sessions table and a messages table. Parts are stored as JSON in a double-wrapped format: [{ "type": "text", "data": { "text": "..." } }]. This is the trickiest format and the main reason this tool exists.

Testing

# Create the OpenCode SQLite test fixture
npm run test:fixtures

# Run all 34 tests (covers all 6 conversion directions + edge cases)
npm test

# Watch mode
npm run test:watch

Tests use 3 hand-crafted mini-sessions (one per harness) that include text, thinking, tool_call, and tool_result content parts. Each test reads from a fixture, writes to a target format, then reads back to verify roundtrip integrity.

Known Limitations

  • Encrypted Codex reasoning (encrypted_content) cannot be decrypted — only summary text is preserved
  • Claude sub-agent sessions (Task tool spawning child sessions) are not followed — only the parent conversation converts
  • Token usage is approximate when cross-converting (different tokenizers per provider)
  • OpenCode must be restarted to discover newly imported sessions
  • System messages are dropped for Claude (not stored in JSONL) or converted to user messages for Codex

Comparison with Alternatives

Feature session-convert agent-convert ctxmv
Claude Code read/write ✅ ✅ ✅
Codex CLI read/write ✅ ✅ ✅
OpenCode read/write ✅ ❌ ❌
MCP interface ✅ ❌ ❌
SQLite support ✅ ❌ ❌
Session preview ✅ ❌ ❌
Path-based conversion ✅ ✅ ❌

Keywords

mcp server, claude code session export, codex cli session migration, opencode session convert, ai coding agent session transfer, claude to codex, claude to opencode, codex to claude, opencode to claude, session migration tool, coding agent context switch, mcp tool, model context protocol, session converter, ai agent session backup, claude jsonl, codex rollout jsonl, opencode sqlite, coding session portability

License

MIT

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选