gpt-control
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., "@gpt-controlrequest an independent review of my latest commit"
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.
GPT-Control
A small cross-model review broker for coding agents.
GPT-Control lets one agent request, track, and verify an independent review from another model. Every submission receives a real run_id; every provider lineage receives a separate conversation_id. Reviews return structured findings, attachment hashes, and a provenance receipt.
An unofficial, community-maintained project. It is not affiliated with OpenAI.
Current status
Experimental. The local contracts are tested, and Chrome Bridge plus Codex are exercised through private live checks. Browser markup can change, provider availability can change, and provider output remains advisory.
Why this exists
The valuable part of a second opinion is the review protocol:
Send only an explicit, bounded file set.
Know which exact submission a result belongs to.
Continue the right provider conversation without conflating it with a run.
Receive findings with file and line evidence.
Record prompt, attachment, and result hashes.
Keep provider transport details behind one small tool surface.
Tools
Tool | Capability | Approval |
| Structured independent review with findings, manifest, and receipt | write |
| Start or continue a provider conversation | write |
| Status, wait, or result for one exact run | read |
| Cancel an active in-process run | write |
| Close local conversation state and owned browser tabs | write |
| Chrome Bridge image generation and iteration | write |
| Transport readiness and focus-safety state | read |
The same core tools are available through the bundled MCP server.
Transports
Transport | Use | Focus behavior | Confirmation |
| Signed-in ChatGPT, conversations, images | Inactive task-owned tab | None |
| Official Codex SDK, structured review, conversations | No browser | None after Codex authentication |
| Official Responses API, structured review, conversations | No browser |
|
| Explicit legacy browser fallback | May take focus |
|
| Explicit legacy API fallback | No browser |
|
Default routing is focus-safe:
Use Chrome Bridge when it is installed and ready.
If Chrome Bridge is installed but leased or unavailable, return a retryable error. GPT-Control does not launch Oracle.
If Chrome Bridge is not installed, use Codex when available.
Oracle browser mode is never selected automatically.
Conversation and run model
Conversation
conversation_id
provider
provider conversation/thread/session id
workspace root
Run
run_id
conversation_id
prompt hash
attachment manifest
status
exact provider result id
result hash
receiptPass conversation_id to gpt_chat or gpt_consult for a follow-up. Pass run_id to gpt_run for status, wait, or result. Per-conversation locking prevents two submissions from interleaving.
Records live under ~/.gpt-control/ by default. Override the root with GPT_CONTROL_HOME.
Attachment boundary
By default, attachments must be regular files under the current workspace after symlink resolution. GPT-Control:
resolves every path with
realpath();rejects directories and special files;
caps file count and aggregate bytes;
blocks obvious credential and private-key paths;
hashes every file with SHA-256;
returns the exact transmission manifest.
Outside-workspace and sensitive-file transmission require separate explicit flags.
What leaves your machine
Prompts and approved attachments are sent to the selected provider. Chrome Bridge uploads files to the signed-in browser session. Codex reads the approved local paths in a read-only sandbox. Responses sends file content in the API request.
Closing a GPT-Control conversation closes local state and wrapper-owned browser tabs. It does not delete provider-side conversations, history, memories, or uploaded files.
Structured review output
gpt_consult returns this contract:
{
"verdict": "request_changes",
"summary": "The migration is not rollback-safe.",
"findings": [
{
"severity": "high",
"claim": "The old schema version is discarded before mutation.",
"evidence": {
"file": "src/migrate.ts",
"lineStart": 81,
"lineEnd": 104
},
"confidence": 0.92,
"remediation": "Persist the old schema version before mutation."
}
],
"openQuestions": []
}Each result also includes a receipt with provider, model, timestamps, provider identifiers, prompt hash, attachment hashes, and result hash.
Install
Oh My Pi
omp install github:wolfiesch/gpt-controlPi
git clone https://github.com/wolfiesch/gpt-control.git
cd gpt-control && bun install
ln -s "$PWD/src/index.ts" ~/.pi/agent/extensions/gpt-control.tsMCP
{
"mcpServers": {
"gpt-control": {
"command": "bun",
"args": ["/path/to/gpt-control/src/mcp.ts"]
}
}
}Chrome Bridge
git clone https://github.com/wolfiesch/chrome-bridge.git
cd chrome-bridge && ./setup.sh
chrome-bridge readyOfficial transports
npm i -g @openai/codex
codex loginFor Responses, provide OPENAI_API_KEY through your normal secret-injection path and pass api_confirmed=true on each paid request.
Configuration
Variable | Purpose |
| Local conversations, runs, locks, and generated artifacts |
| Full Chrome Bridge client command |
| Chrome Bridge checkout containing |
| Full Oracle CLI command |
| Python used to run the bridge client |
| Browser answer poll interval, default 2000 |
| Bridge readiness budget, default 10000 |
| Responses model, default |
Credits
Oracle by Peter Steinberger is the explicit legacy fallback.
Kyle McCleary shared the Oracle fork and browser/image workflow that prompted the first version of this wrapper.
Chrome Bridge provides the focus-safe signed-in browser transport.
The official Codex and Responses adapters use OpenAI's published SDKs.
Development
bun install
bun run check
bun testLicense
MIT
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
Deterministic AI code review, with an audit record. Governance inside the agent loop.
Independent AI-agent reviews: trust checks, evidence scorecards, incident registry, recommendations.
Codex run receipts your reviewer can trust.
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/wolfiesch/gpt-control'
If you have feedback or need assistance with the MCP directory API, please join our Discord server