This plugin is currently in Preview.

Datadog Pi Plugin

Query your Datadog data directly from the Pi coding agent using natural language. Ask about logs, metrics, traces, dashboards, monitors, and more.

What you need

• A Datadog account
• Pi installed (npm i -g @earendil-works/pi-coding-agent)

Why this plugin looks different

Pi does not have built-in MCP support — it is intentionally minimal. This plugin ships as a Pi extension that talks directly to the Datadog MCP server, using the official @modelcontextprotocol/sdk for the wire protocol and OAuth. The extension registers four tools the agent can use:

• datadog — proxy tool the agent uses to list and invoke Datadog MCP tools on demand (token-efficient: one tool definition regardless of how many Datadog tools exist server-side)
• ddsetup, ddconfig, ddtoolsets — manage the connection (site, toolsets, troubleshooting)

Getting started

1. Install the plugin with Pi's package manager:

pi install npm:@datadog/pi-plugin

This adds the package to Pi's global settings and installs the plugin plus its runtime dependencies (including the MCP SDK) into Pi's npm package cache. Restart Pi after install.

2. Start Pi in your project:

pi

3. Configure your Datadog site by asking any Datadog question — the agent will run ddsetup automatically — or invoke it directly:

Run the ddsetup tool with site us1

4. Sign in to Datadog when the agent next calls the datadog tool. Your browser will open to the Datadog sign-in page; once you authorize, control returns to Pi and your tokens are cached globally at ~/.pi/agent/datadog/datadog-oauth/<domain>/. The plugin creates that cache with owner-only permissions. Sign-in is shared across all your projects (per domain), so you authorize once. Subsequent calls reuse the cached tokens and refresh automatically when they expire.

Using the plugin

Just ask the agent anything about your Datadog data:

Show me error logs from the last hour

What monitors are currently alerting?

Find traces for service "api-gateway" with latency > 500ms

List my dashboards

The agent will call the datadog tool with { "list": true } first to discover available tools, then invoke the right one with arguments.

Can't connect?

Never connected before? Tell the agent to run ddsetup. It will configure the Datadog MCP domain globally at ~/.pi/agent/datadog/datadog.json.

Was working before but stopped? Tell the agent to run ddconfig with the troubleshoot action. The most common cause is an expired sign-in — call the datadog tool with { "list": true } to trigger a fresh sign-in flow.

Changing settings

The plugin provides tools the agent can use to manage configuration:

• ddconfig — change your Datadog site or view connection details
• ddtoolsets — enable or disable groups of tools

Advanced usage

API key authentication (for headless / SSH / CI)

If your terminal can't open a browser, you can skip OAuth by setting both Datadog credentials in your shell before starting Pi:

export DD_API_KEY=your-api-key
export DD_APPLICATION_KEY=your-application-key
pi

When both are set, the plugin uses them as request headers and skips the OAuth flow entirely. Run ddconfig to see which auth mode is active in your current session.

Callback port

By default OAuth callbacks come back on http://localhost:19876/callback. Override the port with DD_OAUTH_CALLBACK_PORT if 19876 conflicts with another tool.

Configuration storage and per-project overrides

pi install npm:@datadog/pi-plugin writes the package entry to Pi's global settings (~/.pi/agent/settings.json) and installs the npm package under Pi's package cache (~/.pi/agent/npm/). By default the plugin keeps its config (datadog.json) and OAuth tokens (datadog-oauth/<domain>/) under ~/.pi/agent/datadog/, so your setup and sign-in follow you across every project — you configure and authorize once. The agent dir honors PI_CODING_AGENT_DIR if you've overridden it.

If a single repo needs a different Datadog site than your global default, create a project override: run ddsetup with scope project (or place a .pi/datadog.json in the repo). When that file is present it wins for that directory, and ddconfig/ddtoolsets changes made there stay project-local. OAuth tokens remain global per domain regardless of scope, so a project override never forces a re-login.

Reloading after changes

After updates (pi install npm:@datadog/pi-plugin@latest or local edits), run /reload inside Pi to pick up the changes without a full restart.

Good to know

• Authentication is via OAuth 2.1 + PKCE by default. The MCP SDK handles RFC 9728 discovery, RFC 7591 dynamic client registration, the callback flow, and token refresh.
• Pi does not currently expose a secure extension credential store, so OAuth persistence is file-backed. The plugin stores OAuth artifacts under ~/.pi/agent/datadog/datadog-oauth/, creates OAuth directories with 0700 permissions, and writes artifact files with 0600 permissions.
• No Datadog credentials are sent to the AI model provider — they are only sent in Datadog MCP HTTP request headers.

Support

• Datadog MCP Server Documentation
• Pi Documentation

Legal

See the NOTICE and LICENSE-3rdparty.csv files included with this plugin.

For details on how Datadog handles your data, see the Datadog Privacy Policy.