Feature Requests

Subject: MCP server OAuth rejects the Claude app's client_id ("Claude-Cowork" — invalid_client) Hi Render team, Reporting a gap in your MCP server's OAuth client allowlist. What happens Adding https://mcp.render.com/mcp as a custom connector in the Claude desktop/web app sends the browser to: https://api.render.com/v1/oauth/authorize ?response_type=code &client_id=Claude-Cowork &redirect_uri=https%3A%2F% 2Fclaude.ai %2Fapi%2Fmcp%2Fauth_callback &code_challenge_method=S256... Your authorize endpoint returns, before any consent screen: {"error":"invalid_client","error_description":"unknown or inactive client"} Diagnosis Your docs register specific OAuth client IDs — "claude" for Claude Code, "codex" for Codex, "cursor" for Cursor — and note OAuth is available for those three tools. The Claude app is a separate surface from the Claude Code CLI and sends client_id "Claude-Cowork", which isn't on that allowlist. There's no dynamic client registration to fall back on, and the client_id is set by the app, so there's nothing configurable at my end. Confirmed working for contrast claude mcp add --transport http --client-id claude -s user render \ https://mcp.render.com/mcp Registers fine and reports "Needs authentication" as expected. The CLI path is unaffected — this is specific to the app's client ID. What I'm asking Can "Claude-Cowork" be registered (redirect_uri https://claude.ai/api/mcp/auth_callback ) so the Claude app connector works the way the Claude Code one does? If not, please confirm API-key auth via the Authorization: Bearer header is the only supported route for the Claude app, and consider noting that in the MCP docs — the current page reads as though any Claude client can use OAuth. Context: Render + Neon hosting a Node/Postgres app; I'd like agent access to deploys and logs from the Claude app, not just the CLI. Thanks, Nic Keating

For financial services products (UK is a FinTech hub) many agreements require that data be located within the UK.

Currently, there’s no way to securely use an API key while controlling my web application. I’m developing an API to monitor the server's boot status and attempted to use the List Events feature. However, I can't keep the API key completely safe due to platform limitations. It would be beneficial to have the option to restrict the API key's access to specific actions, like only allowing it to use the List Events API.

My bank card does not allow international payments, and Render currently only accepts debit/credit cards processed in USD. This makes it difficult to upgrade from countries where international transactions are restricted. It would be very helpful if additional payment options (such as local payment methods like UPI or alternative billing options) were supported

I run a monorepo with several frontends and a backend. My backend slugs are significantly bloated because all my code is delivered to the server instead of just the backend code. It would be great to be able to specify paths to trim (not just ignore for redeploys) before a build took place. This feature exists in heroku: https://devcenter.heroku.com/articles/slug-compiler#ignoring-files-with-slugignore I've used it there and is significantly improves build times for monorepos. Additionally there is a large security benefit, since a breach of a frontend server wouldn't expose all my code. Thanks!

For big monorepos or repos with a lot of assets it would be nice to be able to clone with the --filter flag. This would prevent downloading a lot of optional assets. In example, in our case we are developing video games with the Unity Engine. Our server and shared code live in the same repo (I won't enter the reasons here in detail, but think 1 PR = 1 feature, etc.). However each time Render clones the repo, it pulls all the video game assets (being LFS or not) which is costly in bandwith, slower and also can cost us if we are using LFS + Github

My PostgreSQL database (dpg-d98iv06rnols73fdbr4g-a) repeatedly changes from Available to Unavailable every few minutes. The logs show repeated archive-push failures with exit code 82, although the database occasionally recovers temporarily. My production Django application contains important pharmacy data, and the database becomes unavailable, causing my application to stop working and scheduled jobs (such as nightly emails) to fail. Could you investigate why the managed PostgreSQL instance is repeatedly becoming unavailable and help stabilize or recover it?

How a WhatsApp bot works