codex-session-gateway
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@codex-session-gatewaycontinue the Codex session on my remote dev host in the api repo"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
codex-session-gateway
Self-hosted MCP gateway for securely connecting ChatGPT and Claude to existing Codex sessions across local and remote development hosts.
Early development — scaffold only (0.0.0). There is no runnable gateway, agent, dashboard, MCP endpoint, authentication, or Codex integration. Do not deploy this scaffold as a service. The development toolchain and scaffold are verified locally; see bootstrap status. GitHub Actions is configured but not externally verified.
Project status
Capability | Status | What exists |
Monorepo source and development configuration | Implemented | Metadata exports, six smoke tests, build and quality checks verified locally |
Architecture, examples and roadmap | Implemented | Design documents; examples are not executable configuration |
CI, CodeQL, dependency review and Dependabot definitions | Implemented | Configured but not externally verified; local development is independent of Actions |
MCP gateway and browser connectors | Planned | No SDK or server installed |
OAuth and agent credentials | Planned | No authentication code |
Outbound agent transport and Codex adapter | Planned | No network connections or session access |
Dashboard, persistence and deployment | Planned | Documentation only |
Generic MCP shell or arbitrary remote filesystem access | Not supported | Excluded from the intended interface |
Implemented means source or documentation is present, not production ready. Experimental means executable behavior under evaluation; no runtime feature currently has this status. Planned means design intent without implementation. Not supported means outside the supported scope.
Related MCP server: opencode-chatgpt-bridge
Problem and intended use cases
Development work can be spread across laptops, workstations and remote hosts. A browser assistant needs an explicit, authorized way to select the correct host, workspace and Codex thread without treating every machine as one shared shell.
Planned use cases include inspecting an existing thread, continuing work in an allowlisted repository, observing a run, and requesting interruption. These are project goals, not available integrations. Existing-session compatibility must be verified against supported Codex versions before any such claim is made.
Planned architecture
flowchart LR
Clients[ChatGPT Web / Claude Web] -->|Planned: MCP over HTTPS with OAuth| Gateway[Public gateway]
Agent[Node agent on development host] -->|Planned: outbound WSS only| Gateway
Agent -->|Planned: local adapter| Codex[Codex App Server]
Codex --> Workspace[Allowlisted repository on the same host]The node agent initiates the WSS connection; authorized requests would travel over that established connection. Codex App Server would run on the machine containing the repository. Targets would include host, workspace and thread identity, with one active mutation per thread. The gateway would have no direct access to remote filesystems. None of these connections exists in the scaffold.
Monorepo
Path | Responsibility |
| Public gateway metadata; service planned |
| Host agent metadata; service planned |
| Documentation and package manifest only |
| Metadata and Zod dependency reserved for future runtime contracts |
| Metadata; local Codex adapter planned |
| Metadata; shared authorization primitives planned |
| Metadata; shared utilities may be introduced when needed |
| Nonfunctional, sanitized configuration examples |
| Caddy, Docker and systemd planning notes only |
| Architecture, operations, security and decision records |
| Local documentation and preventive secret checks |
| Community templates and automation definitions |
All packages use @codex-session-gateway/*, ESM and version 0.0.0. They
remain private in package manifests to prevent accidental package publishing;
the intended GitHub repository is public. Changesets can version private
packages.
Development quick start
Use an unprivileged account, Node.js 24.20.0 LTS (pinned in .node-version)
and pnpm 10.34.5 (pinned in package.json). See the verified
toolchain installation method. The generated
pnpm-lock.yaml is committed; frozen installation is the normal local and CI
workflow.
Run the entire suite locally from the repository root:
pnpm install --frozen-lockfile
pnpm lint
pnpm format:check
pnpm typecheck
pnpm test
pnpm build
pnpm check:docs
pnpm check:secrets
pnpm audit --audit-level highBuild emits only metadata modules and declarations into ignored dist/ folders.
There is no start command and no development HTTP server. See
development for formatting and Changesets usage.
Future runtime requirements
Planned: a public HTTPS origin, an OAuth design compatible with the selected clients, outbound WSS access from development hosts, and a supported local Codex App Server on each repository host. Client account eligibility, protocol versions, operating systems, credential storage and persistent state are not yet validated. No DNS, certificate, firewall, SSH or service configuration is supplied or changed.
Security model
Planned: deny by default, least privilege, explicit target authorization, workspace allowlists enforced locally, scoped revocable agent credentials and per-thread mutation locks. Repository content and model output are untrusted. Credentials, raw transcripts and sensitive tool payloads must not enter normal logs. A compromised gateway must not bypass the agent's local policy. There is no generic shell MCP tool. No runtime enforcement exists yet.
Read the security model and threat model. The local secret checker is preventive and heuristic; it cannot certify that a repository contains no secrets.
Documentation
Start with getting started and architecture. Other entry points are authentication, MCP tools, transport requirements, configuration, ChatGPT, Claude, operations and troubleshooting.
Roadmap and contributing
ROADMAP identifies future milestones without delivery dates. Use
CONTRIBUTING for branch workflow, checks and Conventional
Commits, and SUPPORT for questions. Maintainer: Alberto Lopez -
Alcybercloud.it (GitHub owner: peronslayer). Repository:
https://github.com/peronslayer/codex-session-gateway. Source publication and
local scaffold verification do not constitute a release or validation of the
planned runtime integrations.
Vulnerability reporting
Do not open public issues for vulnerabilities. Follow SECURITY. A verified private reporting channel is a bootstrap requirement; do not post sensitive details while that channel is pending.
License and independence
Copyright 2026 Alberto Lopez - Alcybercloud.it. Code is licensed under Apache License 2.0. The Code of Conduct uses Contributor Covenant 2.1 with its upstream attribution preserved.
This is an independent open-source project and is not affiliated with, endorsed by, or sponsored by OpenAI, Anthropic, Microsoft, or Visual Studio Code.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
No tool schema history has been recorded yet.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
- QuallaaOAuthcom.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Persistent memory and cross-session learning for AI coding assistants (hosted remote MCP).
Enable secure connectivity between Sentry issues and debugging data, and LLM clients, using a Model Context Protocol (MCP) server.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceExposes local OpenCode instances as remote MCP servers for Claude and ChatGPT, enabling terminal access, session management, and interactive human-in-the-loop workflows. It simplifies deployment for local machines using Cloudflare Tunnels to provide secure public connectivity and OAuth support.-
- AlicenseNot gradedqualityBmaintenanceA local MCP bridge that lets ChatGPT control opencode sessions for code modification, file reading, and repository management on your own computer.1MIT
- AlicenseAqualityBmaintenanceBridges OpenAI Codex CLI to any MCP client, allowing headless Codex sessions via tools like codex and codex-reply.2261MIT
- AlicenseNot gradedqualityBmaintenanceLocal MCP bridge enabling ChatGPT web to access approved local files and execute tasks via local Codex.6MIT
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/PeronSlayer/codex-session-gateway'
If you have feedback or need assistance with the MCP directory API, please join our Discord server