Wrapper
Provides tools for using a ChatGPT (OpenAI) browser session to queue and run coding, review, analysis, or image tasks. Supports submitting tasks, polling status and event history, cancelling work, retrieving response artifacts, and inspecting metrics and errors. It uses the user's ChatGPT account/session rather than the OpenAI API.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
ChatGPT Wrapper
I ran out of Codex usage, so I literally wrapped ChatGPT.
Connect your own ChatGPT account to coding agents through MCP. Wrapper runs locally, queues tasks in SQLite and sends requests over HTTP. Use your own ChatGPT web session; no automated browser runtime or Chromium installation is needed.
Current source status
Version 0.7.0-dev.2 is unpublished. One live public text task completed through production MCP, web HTTP preparation/message submission, v1 SSE decoding and SQLite persistence in 4.767 seconds. The successful session required a fresh cookie and matching short-lived Sentinel bundle. Cookie login and refresh are fixture-covered and passed one isolated live check with a matching captured cookie, User-Agent and model; no message was submitted, and remote logout remains unverified. Login without a captured User-Agent received HTTP 403 from ChatGPT's web session endpoint. A photo-stage file registration, signed upload and acknowledgement completed (HTTP 200, 201 and 200), but the conversation POST returned HTTP 403 while the captured requirements had about 65 seconds remaining by their recorded expiry. No retry was attempted; this does not establish why the provider rejected it. Model-side file or image recognition remains unverified. Search, image generation/analysis/editing and conversation continuation lack live capability evidence. See the capability evidence matrix and HTTP transport details.
Published npm 0.6.0 predates this HTTP-only migration. Build the current source using source installation, then configure web authentication:
# First-time setup uses an existing signed-in ChatGPT browser session.
node dist/cli.js auth login --cookie-stdin --model "MODEL_SLUG" < /absolute/path/private-cookie.txt
node dist/cli.js auth-status
node dist/cli.js statusauth verifies the existing web session. The existing auth --token-stdin path accepts a bearer credential through standard input. auth login reads a full ChatGPT browser-session Cookie header from standard input; it does not accept an email/password or open a browser. Replace MODEL_SLUG with the backend model slug from your own session. If required by your session, pass --user-agent with the value captured from that same authorized ChatGPT request; it is stored and reused by refresh/logout. The command validates the cookie through ChatGPT's web session and identity endpoints, then stores the cookie and bearer in the private session file. Keep any file or pipe used for cookie input private and never paste credentials into chat or command arguments.
auth refresh obtains a fresh bearer using the stored cookie and updates the private session file. It does not refresh Sentinel requirements, proof, or conduit values; a task may still need a fresh authorized capture. auth-status reports local configuration only and does not test whether ChatGPT will accept a message. See HTTP transport details for logout and session limitations.
Related MCP server: AKIMCP
What it does
MCP task submission, status, waiting, cancellation and result retrieval.
Durable SQLite tasks and batches, progress snapshots and local diagnostics.
Direct HTTP requests using a private ChatGPT web session.
Selected file and image transfer has a live registration/upload/acknowledgement check. Search and image generation/analysis/editing have candidate HTTP paths with fixture coverage; model-side behavior remains unverified.
Backend model-slug selection and conversation-parent verification before follow-ups.
Explicit failures without automatic prompt retries or browser fallback.
flowchart LR
A[Coding agent] --> B[MCP server]
B --> C[Local SQLite queue]
C --> D[HTTP worker]
D --> E[ChatGPT web endpoint]
E --> F[Validated result]
F --> AWrapper returns results to your MCP client; the client handles planning, source files, applying changes and reviewing generated output. Wrapper does not execute generated code or edit your repository.
Useful commands
After building the source:
node dist/cli.js doctor
node dist/cli.js status
node dist/cli.js watch <task-id>
node dist/cli.js stopauth-status shows safe local web credential configuration without tokens or a network request. node dist/cli.js auth logout signs out the stored ChatGPT session wherever its cookies are used and removes local credentials only after remote signout succeeds. auth logout --local-only removes local credentials without contacting ChatGPT. get_status reports provider chatgpt-web-api and HTTP transport. api-import --har /absolute/path/capture.har imports HAR; api-import --capture /absolute/path/capture.txt imports Network text; and api-import --capture /absolute/path/capture.proxymanlogv2 imports a Proxyman archive. Keep captures private; fresh captures may be needed as requirements expire.
Treat HAR captures, session files and local tokens as credentials. Prompts and results remain in the private task store; prompts are sent to ChatGPT. Account limits and verification requirements still apply. Security and privacy.
User-supplied image URLs require exact host entries in ALLOWED_IMAGE_HOSTS. Signed URLs returned by the authenticated ChatGPT file endpoints accept the observed one-label *.oaiusercontent.com hosts by default; exact per-direction overrides are available for other hosts. HTTPS on the standard port and redirect refusal apply to both flows. ChatGPT credentials are never sent to asset hosts.
Documentation
Report problems with a minimal public reproduction; never attach credentials, HAR captures or private task data. Open an issue · Report a security issue privately.
Licensed under GPL-3.0.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Headless task management for bots, apps and AI agents. Remote MCP server, OAuth or access token.
Shared memory for your AI tools: tasks, notes and context every tool can read and write over MCP.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables ChatGPT (or any MCP client) to delegate coding tasks to a local Hermes-backed agent with async job management, supporting read-only investigation, implementation, and continuation of sessions via secure MCP tunnel.1MIT
- FlicenseNot gradedqualityAmaintenanceEnables web-based AI chat clients to securely operate a local machine through one OAuth-gated MCP endpoint, exposing governed tools for file read/write/edit, whitelisted shell commands, search, Git, SQLite, browser automation, DevTools, background tasks, localhost fetches, clipboard, notifications, ports, and Postman. It lets users build and edit projects, inspect databases and local APIs, drive browser workflows, and manage long-running jobs from a browser or mobile device without granting unrestricted shell access.-
- AlicenseNot gradedqualityAmaintenanceEnables AI clients such as Claude, Cursor, VS Code, and Windsurf to manage workspaces, projects, tasks, and AI agents through natural language, using a hosted OAuth-authenticated endpoint or a local stdio package. It also ships an OpenAPI codegen tool for generating MCP tools from any OpenAPI spec.2,530 npmMIT
- FlicenseNot gradedqualityAmaintenanceEnables MCP clients to drive Codex Desktop and codex-app-server as a durable asynchronous worker for long-running tasks, with polling, approvals, Plan Mode workflows, retry safety, diagnostics, and final reporting. It also provides recovery, queueing, and repair tools for reliable local automation.-