MCP

claude-codex-session-link: An MCP Server for Reading Claude Code Sessions in Codex

If you want to find and read local Claude Code sessions directly from OpenAI Codex, this stdio MCP server is a good fit; but it requires Node.js 22+, locally present Claude Code session files, and true send-back actions go online and consume Claude usage.

Official source · GitHub - Dayc-l/claude-codex-session-link

Install

git clone https://github.com/Dayc-l/claude-codex-session-link.git && cd claude-codex-session-link && node.exe cli.mjs list --limit 10

claude-codex-session-link is an MCP server that connects local Claude Code sessions to OpenAI Codex. If you want to search and read locally saved Claude Code conversation logs directly from Codex, it can do that; but note: true "follow-up asking / sending back" requires an authenticated Claude Code CLI, goes online, and consumes Claude usage.

Who It's For

  • Developers who use both OpenAI Codex and Claude Code and want to find and read locally saved Claude Code conversations from within Codex.
  • Users on Windows (the official documentation states the local reader has been verified on Windows).
  • Not suitable for: users without local Claude Code session files; environments that cannot meet the Node.js 22 or newer requirement; users who want purely read-only behavior but also want to trigger send-back without an authenticated Claude Code CLI. Compatibility on other operating systems is not stated in the evidence and cannot be confirmed.

First Try in 3 Minutes

Prerequisites: Node.js 22 or newer installed locally, and locally present Claude Code session files.

Main path (officially supported, no npm install needed):

git clone https://github.com/Dayc-l/claude-codex-session-link.git
cd claude-codex-session-link
node.exe cli.mjs list --limit 10

list --limit 10 is a low-risk read-only call: it lists recent local sessions. The success signal is the command outputting a session list (up to 10 entries).

Hooking into Codex: in the .codex/config.toml of a project you trust, add:

[mcp_servers.claude_session_link]
command = "node.exe"
args = ["C:\\Tools\\claude-codex-session-link\\cli.mjs", "mcp"]

(Replace the path with where you actually cloned it.) The server runs over stdio MCP / JSON-RPC, with no HTTP service and no runtime npm dependencies.

Note the distinction between two kinds of operations: claude_sessions, claude_read_session, and claude_session_updates are local read operations; claude_ask_session involves real send-back, accesses the network, consumes Claude usage, and requires an authenticated Claude Code CLI. The official documentation does not state whether this operation has an undo path, so confirm for yourself before executing. A read-only label only means no write operations are performed; it does not mean no network access or no privacy implications.

Use Cases

  1. Searching local Claude Code sessions from within Codex: the scenario is that you can't remember the title or time of a conversation. Call claude_sessions (lists and searches local conversations, read-only, returns a list for the user to choose from). Have the model call this tool in Codex and you'll see matching session candidates; the specific search parameters are not individually listed in the official evidence, so you can start with a parameterless list call to confirm.

  2. Reading the full content and increments of a session: the scenario is that after finding the target session you want to see what was discussed. First use claude_sessions to get the session, then call claude_read_session to read its content; if you want to know about new changes after the session, call claude_session_updates to get updates. All of these are local reads; you can be confident up to this point.

The details of the claude_ask_session send-back flow are not elaborated in the official evidence; do not treat it as a read-only attempt without confirmation.

Usage Boundaries

  • Read/write nature: the three read-type tools are read-only local reads; claude_ask_session is a real send-back that goes online and consumes Claude usage, provided the local Claude Code CLI is authenticated.
  • Local requirements: Claude Code session files must exist; Node.js 22+.
  • Side effects and confirmation points: the send-back action incurs consumption on the Claude side, and the official documentation gives no undo path; writing the configuration to .codex/config.toml is a local state modification.
  • Official limitations: the local reader has only been verified on Windows; no runtime npm dependencies, no HTTP service.
  • Not yet confirmed: non-Windows compatibility, the specific parameters and confirmation flow of claude_ask_session, and the full parameter lists of the read tools — none have official evidence, so verify them yourself in the repository before use.