Allows an authorized ChatGPT Work Cloud worker to look up context from Spaces/Meeting sources, perform bounded edits to authorized Spaces Pages, and save selected transcripts as Spaces Pages through queued bridge requests.
Click on "Deploy 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., "@Spaces Context Bridge Look up what we decided about the pricing change in my Spaces"
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.
0.1.5 — open-source and self-hosted. A self-hosted, single-owner MCP relay for asking a ChatGPT Work Cloud worker to look up context, edit an authorized Page, or save a selected transcript as a Page.
The bridge has no direct Spaces access. It stores requests in Cloudflare D1; a separately authorized ChatGPT worker uses the owner's available tools. OAuth authentication is not worker-event activation. A queued request is not a completed operation. There is no automatic transcript sync.
Each installation runs in its owner’s Cloudflare account with its own database, connection codes and authorized ChatGPT worker. Maintainers do not operate a shared notes service.
Fresh Cloudflare CLI deployment, hosted OAuth/protocol checks, and all three synthetic bridge operations through T3/Codex → Executor → bridge → ChatGPT cloud event task → actual Spaces Pages have passed. Lookup sources, the bounded Page edit and the verbatim transcript Page were independently verified. No local daemon was needed for this cloud route. Additional clients and the complete public deployment-button flow still need verification; automatic transcript sync is not implemented. See evidence and remaining gates. Licensed under MIT, copyright 2026 Design X Develop; see LICENSE.
Related MCP server: workspace-agent-relay-mcp
lookup_context: enqueue a question for authorized Spaces/Meeting sources; retrieve sources with get_result.
request_document_edit: request a bounded edit to an exact Page ID or URL, within the owner's configured scope.
save_transcript: enqueue a selected transcript and destination Space; creates a Page, not a native ChatGPT Meetings recording.
bridge_status: report role and subscription state. worker_connected: true alone does not prove the worker processes requests.
Claude app and Claude Code are agent clients. See T3 Code, OpenCode CLI, and optional Executor for additional setups. Public users can keep direct individual-client connections; Executor is optional. For the release owner’s personal setup, consolidate agent access into Executor only after verification. Keep the ChatGPT worker direct because it needs separate worker permissions and events. The ChatGPT worker is a different role with queue-processing tools. Official Superwhisper MCP runs locally and is separate from this remote relay. “Private” means owner-operated and access-controlled, not local-only processing or end-to-end encryption.
Use the setup prompt for your chosen agent or MCP host to connect an existing endpoint or plan a deployment on your chosen backend. Cloudflare Worker + D1 is the currently implemented backend; other hosts require adapters and acceptance testing. An MCP client/router is separate from the backend server. The original ChatGPT-mediated Spaces lookup/edit/save workflow stays the same.
The button deploys into your own Cloudflare account. The complete browser-button flow is still unverified; the documented CLI flow has passed a clean installation with fresh resources. Human authentication, worker activation and acceptance testing remain part of setup. Cloudflare documents this official button format and resource provisioning.
Follow fresh deployment using a new Worker and D1 database.
Connect your agent client and ChatGPT worker using client setup; enter connection codes only on the consent page.
Set the owner scope in the worker prompt, then activate request.created.
Run the disposable acceptance checks. Keep real transcripts out until they pass.
No publisher account ID, database ID, personal origin, Page ID, or connection code should be reused. The repository URL above identifies source code, not a running service.
An agent calls the HTTPS /mcp endpoint → the Worker authenticates and queues a request in D1 → a signed event wakes the ChatGPT worker → it claims the request and uses authorized tools → it records a verified result → the original client polls get_result.
The Worker stores payloads and results, including submitted transcript text. Event notifications contain a request ID and kind; the worker subsequently retrieves the payload. Returned context passes through D1 and the requesting model provider. See data flow and operations and security.
Cloudflare Workers and D1 have free allowances, but this project has no measured monthly cost. Workers Paid starts with a $5 monthly subscription plus applicable usage. D1 charges by rows read/written and storage beyond included limits. Check Workers pricing and D1 pricing before deployment. ChatGPT/Claude plans and usage, Superwhisper features, and optional domains are separate. No OpenAI API key is required by this relay's source; that does not make ChatGPT Work execution free.
One deployment represents one owner. It is not a tenant-isolated service for unrelated people. Page authorization is enforced by worker instructions and source permissions, not by a server-side Page allowlist. Cleanup currently removes requests by creation age after seven days when cron runs. Retries, leases, and request idempotency do not guarantee exactly-once Page edits.
Research and compatibility covers official product facts and access limits. Optional companion design describes a future opt-in transcript sender; it is not implemented. Contributing explains local checks. This project is independent of OpenAI, Anthropic, Cloudflare, Superwhisper, and Granola.
Use GitHub Issues for setup questions, reproducible bugs and feature requests. Include the version, client, setup step and a synthetic reproduction; omit credentials, transcripts, database exports and private links.
Report security vulnerabilities through GitHub private vulnerability reporting, as described in SECURITY.md.
This server cannot be deployed
Cloudflare Workers MCP server: agent-workflow-engine
Remote streamable-HTTP MCP server running on a single Cloudflare Worker. Your assistant gets live Airbnb, Amazon, Booking.com, Google Flights, Maps and Reddit data, social search on X, Instagram and TikTok, the Meta Ad Library, and image/video generation without any keys. Connect your own accounts to let it send WhatsApp or Telegram messages, work an IMAP inbox, manage Meta Ads campaigns and publish to X and LinkedIn. OAuth 2.1 with PKCE; stored credentials are AES-256-GCM encrypted.
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
