Switching clouds? Get up to $10K in credits + hands-on help.

Apply now
Back to agents

Pi + Render

Deploy and manage your Render apps from Pi.

Pi is a minimal, self-extensible terminal coding agent. It ships no MCP support of its own, so Render's Pi package supplies both halves of the integration: Render's official skills for deploys, Blueprints, debugging, and monitoring, plus hosted Render MCP access through the pi-mcp-adapter extension. With both installed, Pi can inspect services, fetch logs, update environment variables, create supported services, and query Render Postgres from your shell.

Setup contract
Config file
~/.pi/agent/settings.json
MCP install command
pi install https://github.com/render-oss/render-pi-plugin
Skills command
pi install https://github.com/render-oss/render-pi-plugin
Skills directory
~/.pi/agent/git/github.com/render-oss/render-pi-plugin/skills/
Auth
OAuth
Restart required
Yes
Verify
list_workspaces
Agent details
Agent
Pi
Install
npm install -g --ignore-scripts @earendil-works/pi-coding-agent
Start
pi
Skills mode
Agent-invoked

Quick start

  1. Install Pi.

  2. Install the Render Pi package. This installs Render's skills and the Render MCP configuration together:

    To pin the install to a specific release:

    Pi installs the package's runtime dependencies, including its MCP adapter. There's no separate MCP package to install.

  3. Restart Pi so it loads the new extension and skills.

  4. Connect to Render. In a Pi session, run:

    This opens the OAuth flow in your browser. For non-interactive environments, set RENDER_API_KEY instead. See the MCP configuration section below.

  5. Set your Render workspace if you have multiple workspaces.

  6. Deploy a template. Try a prompt like "Deploy this project to Render and show me the plan before applying it."

On a clean first run, the adapter makes one best-effort connection to populate its tool-metadata cache before you have credentials, so Pi reports an HTTP 401 from the Render server once. This is expected. Pi and every Render skill still load normally, and the notice stops after you run /mcp-auth render.

MCP configuration

The Render Pi package configures the Render MCP server programmatically, so there's no MCP config file for you to edit. Pi records the package itself in ~/.pi/agent/settings.json when you run pi install.

MCP setup
Install command
pi install https://github.com/render-oss/render-pi-plugin
Config file
~/.pi/agent/settings.json
OAuth
Supported

Run this in a Pi session:

Your browser opens to sign in with your Render account, and credentials land in your operating system's credential store. No API key is needed. For details, see the MCP server docs.

API key

Set RENDER_API_KEY in the environment before starting Pi:

The package reads the variable when it makes a lazy connection and sends it as a bearer token. Generate a key in your account settings. Use this method for non-interactive environments like CI/CD pipelines, and on headless Linux where OAuth needs an unlocked system credential store.

Because the package owns its own MCP configuration, the Render server stays isolated from any MCP servers you set up yourself. The trade-off is that /mcp setup, /mcp enable, and /mcp disable don't apply to it, and /mcp status reports only what the package configured.

The prompts below can use a combination of MCP tools, Render skills, the Render CLI, and guided Dashboard or API steps.

Example prompts

Start with one of these prompts, then refine the plan with your agent.

Deploy

Deploy this project to Render.

Debug

Check the deploy logs for the last failed deploy and fix the error.

Blueprint

Create staging and production environments using a render.yaml Blueprint.

Database

Create a Render Postgres database and wire DATABASE_URL to my web service.

Render skills

Render's official skills give Pi task-specific procedures for deploys, debugging, monitoring, databases, Blueprints, and more. Pi keeps each skill's name and description in context, loads the full SKILL.md on demand, and registers every skill as a /skill:render-* command.

Install skills

The Render Pi package is the preferred way to install skills, and it installs the Render MCP configuration at the same time:

To get the skills without the MCP tools, run pi config and disable the package's extension while leaving its skills enabled.

Use the standalone skills installer if you don't want the package at all:

Skills catalog

View Render skills on GitHub

Common workflows

Deploy from the terminal

  1. Ask Pi to inspect the project and propose a Render deployment plan.
  2. Review the generated Blueprint or service settings.
  3. Let Pi apply supported changes with MCP or the Render CLI, or have it guide you through Dashboard or API steps.
  4. Ask Pi to watch the first deploy.

Debug a failing deployment

  1. Ask Pi to fetch recent deploy logs.
  2. Have it identify the most likely root cause.
  3. Apply the code or env-var fix, then verify the next deploy.

Add a database

  1. Ask Pi to create a Render Postgres database.
  2. Store the internal connection string as DATABASE_URL.
  3. Run migrations from a pre-deploy command.

Tips for success

  • Restart Pi after installing or updating the package. Extensions and skills load at startup.
  • Invoke a skill directly with /skill:render-deploy when you want a specific procedure instead of letting the model choose.
  • Keep Render config in the repository so Pi can inspect infrastructure changes.
  • Ask for a plan before applying changes when adding services, databases, or workers.

FAQ

Troubleshooting

The Render tools don't appear

Restart Pi. Extensions load at startup, so a pi install in another terminal has no effect on a running session. Then run /mcp-auth render and confirm /mcp status lists the render server.

/mcp setup, /mcp enable, or /mcp disable don't affect Render

That's expected. The package supplies the Render server programmatically rather than through your MCP config file, which keeps it isolated from your own servers. Use pi config to disable the package's extension if you want to turn the Render tools off.

OAuth fails on headless Linux

The OAuth flow writes credentials to the system credential store, which needs an unlocked keyring. Set RENDER_API_KEY before starting Pi instead.

Community

Other coding agents that pair well with Render.