gpt2agent

name	gpt2agent
description	Full ChatGPT Plus/Pro account access via MCP. 25 tools covering chat, agent mode, deep research, image generation, code execution, canvas, memory, custom instructions, conversations, Custom GPTs, and Codex. Reuses $CODEX_HOME/auth.json (or ~/.codex/auth.json) or the manual ~/.gpt2agent/token.json fallback. Use when you need ChatGPT models, web research with citations, DALL-E image gen, Python sandbox execution, or ChatGPT account management.
allowed-tools	["Bash","Read","Write","mcp__gpt2agent__chat","mcp__gpt2agent__agent","mcp__gpt2agent__deep_research","mcp__gpt2agent__deep_research_heavy","mcp__gpt2agent__gpt_chat","mcp__gpt2agent__generate_image","mcp__gpt2agent__get_file_info","mcp__gpt2agent__get_file_download_url","mcp__gpt2agent__code_interpreter","mcp__gpt2agent__canvas_execute","mcp__gpt2agent__account_status","mcp__gpt2agent__list_models","mcp__gpt2agent__list_conversations","mcp__gpt2agent__get_conversation","mcp__gpt2agent__list_tasks","mcp__gpt2agent__list_apps","mcp__gpt2agent__list_custom_gpts","mcp__gpt2agent__memory_list","mcp__gpt2agent__memory_search","mcp__gpt2agent__memory_create_via_chat","mcp__gpt2agent__custom_instructions_get","mcp__gpt2agent__custom_instructions_set","mcp__gpt2agent__list_codex_envs","mcp__gpt2agent__list_codex_tasks","mcp__gpt2agent__codex_task_create"]

gpt2agent — ChatGPT Account Access via MCP

command -v gpt2agent >/dev/null && echo "gpt2agent: installed ($(gpt2agent --version 2>/dev/null || echo 'unknown version'))" || echo "gpt2agent: NOT INSTALLED — run: pipx install gpt2agent"
if test -f "${CODEX_HOME:-$HOME/.codex}/auth.json"; then
  echo "codex token: present"
elif test -f "$HOME/.gpt2agent/token.json"; then
  echo "gpt2agent token: present (manual fallback)"
else
  echo "ChatGPT token: MISSING — run: codex login or gpt2agent setup"
fi

Preconditions

1. gpt2agent installed via pipx. If missing: pipx install gpt2agent
2. Auth token at $CODEX_HOME/auth.json (or ~/.codex/auth.json, from codex login) or ~/.gpt2agent/token.json.
3. MCP server registered in Claude Code. If missing: gpt2agent install
4. Restart Claude Code session after first install to load MCP tools.

If any precondition fails, stop and tell the user the exact fix command.

Quick Reference

Category	Tools
Chat & reasoning	chat, agent, deep_research, deep_research_heavy, gpt_chat
Image & file	generate_image, get_file_info, get_file_download_url
Code execution	code_interpreter, canvas_execute
Account & models	account_status, list_models, list_apps
Conversations	list_conversations, get_conversation, list_tasks
Custom GPTs	list_custom_gpts, gpt_chat
Memory	memory_list, memory_search, memory_create_via_chat
Instructions	custom_instructions_get, custom_instructions_set
Codex	list_codex_envs, list_codex_tasks, codex_task_create

When to Use Which Tool

Need	Tool	Notes
Quick Q&A with a ChatGPT model	chat	Default: gpt-5-3, temporary=True
Chat that needs image gen / code / canvas	chat(temporary=False)	Temporary chats block these features
Multi-step task with browsing + code exec	agent	262K context, autonomous
Web research with citations (30-120s)	deep_research	Costs 1 DR quota
Long-form research report (5-30 min)	deep_research_heavy	Costs 1 DR quota, use background mode
Generate an image	generate_image	DALL-E, requires temporary=False context
Run Python in sandbox	code_interpreter	Requires temporary=False
Live document editing	canvas_execute	Requires temporary=False
Use a Custom GPT	gpt_chat(gizmo_id, prompt)	List IDs with list_custom_gpts
Save something to ChatGPT memory	memory_create_via_chat	Model-initiated write (REST 405 workaround)
Create a Codex coding task	codex_task_create	Auto-resolves environment_id from repo_label

Usage Patterns

Research workflow

1. deep_research("topic")          -- fast, citations included
2. deep_research_heavy("topic")    -- long-form, run in background
3. Read result, summarize for user

For deep_research_heavy, use run_in_background: true in Bash. It takes 5-30 minutes.

Image generation

1. chat("describe what you want", temporary=False)  -- or generate_image directly
2. generate_image("prompt")  -- returns file_id
3. get_file_download_url(file_id)  -- get the URL

Code execution

1. code_interpreter("python code here")  -- runs in ChatGPT sandbox
2. canvas_execute(content="...")  -- live document editing

Both require a non-temporary conversation context.

Account inspection

1. account_status()   -- plan, features, expiry
2. list_models()      -- all available model slugs
3. list_conversations() -- recent chat history

Codex integration

1. list_codex_envs()              -- see available repos
2. list_codex_tasks()             -- recent tasks
3. codex_task_create(repo, desc)  -- spin up a coding task

Quota Management

Deep Research limits and reset timing are reported by the current account and can change.

• deep_research and deep_research_heavy each cost 1 quota.
• Run the bundled deep-research/bin/quota.sh before heavy usage.
• Warn user if remaining quota < 10 before firing deep research.

Configuration

Config file locations (checked in order):

1. ~/.gpt2agent/config.toml
2. ./config.toml (cwd)
3. ~/.config/gpt2agent/config.toml

Key settings:

[server]
host = "127.0.0.1"   # loopback only; HTTP transport is unauthenticated
port = 9000

[models]
chat = "gpt-5-3"
# agent = "agent-mode"
# heavy_dr = "gpt-5-5-pro"

Troubleshooting

Error	Fix
"gpt2agent not installed"	pipx install gpt2agent
"ChatGPT token: MISSING"	codex login or gpt2agent setup
MCP tools not appearing	Restart Claude Code session after gpt2agent install
401 / auth error	Re-run codex login or gpt2agent setup, then restart the MCP server
Image/code/canvas fails	Ensure temporary=False — these features are blocked in temporary chats
DR connector unavailable	Enable Deep Research at chatgpt.com > Settings > Connectors
"memory_add not available"	Use memory_create_via_chat instead (REST POST returns 405)

Detailed Reference

For full parameter docs and return types, see: gpt2agent/gpt2agent/skills/gpt2agent/tools-reference.md

Or inspect tool schemas directly: mcp__gpt2agent__* tools are self-describing via MCP.