Plausible alternative with MCP for AI agents

I used Plausible-style analytics for about a year before I started building Lodd. Good product, clean dashboard, right approach to privacy. The problem was that I kept forgetting to check it. My analytics lived in a browser tab and I worked in Claude Code, and the two never met. I wanted traffic data in the same session as my code, not in a separate window I'd remember to open once a week.

Plausible doesn't have an official MCP server. Community wrappers exist with 4 tools, but that's a narrow surface. If your agent is where you work, and your analytics can't reach your agent, you're still context-switching even if the dashboard is nice. This post compares the two approaches.

Does Plausible have an MCP server?

Plausible has no official MCP server. Two community-built wrappers exist: one from Sentry's team with four tools (get_timeseries, get_breakdown, get_conversions, compare_periods) and one from alexanderop with four tools as well. Both use your Plausible API key and work, but they're maintained by individuals, not the Plausible team, and four tools is a narrow surface for anything beyond quick traffic checks.

No funnels, no realtime, no conversion attribution, no actor analytics. If you need those through an agent, the community wrappers won't cover it.

How does Lodd compare to Plausible?

What can an agent do with Plausible's community MCP?

The Sentry-built wrapper gives your agent four operations: pull a time series of visitors, break down traffic by any dimension (source, page, country), fetch goal conversions, and compare two time periods. That covers the basics. Your agent can check "how's traffic this week vs last week" or "what are my top referrers." For monitoring and quick checks, it works.

What it can't do: build multi-step funnels, check real-time visitor counts, attribute conversions to specific pages or traffic sources, track individual actor journeys, create annotations, or manage trackable links. These aren't things Plausible lacks as a product, they're things the community wrapper doesn't expose. Plausible's own API is broader, but nobody has wrapped the full surface into MCP tools yet.

What can an agent do with Lodd's 42 tools?

Lodd's MCP server covers the full analytics surface: snapshots, time series, funnels with conversion rates between steps, breakdowns by page, source, country, browser, device, and OS. Entry and exit page analysis. Realtime visitor counts. Custom event tracking with counts, individual records, and time series. Conversion attribution showing which pages and sources drive specific events. Actor analytics with per-user timelines and cohort retention. Trackable short links with click tracking. Annotations that embed automatically in time series data.

The responses come back as compact JSON with shortened keys, designed for agent context windows. A full analytics snapshot is roughly 55 tokens.

When should you use Plausible instead?

If you need charts on a wall screen, in board presentations, or in a reporting workflow that requires exported visuals, Plausible has a dashboard and Lodd doesn't. That's a real gap for those specific cases. But most people who think they "need a dashboard" actually need answers to simple questions like "how's traffic this week?" A non-technical teammate can ask Claude Desktop that and get a plain-English summary that's clearer than any chart. The dashboard assumption is worth questioning.

Plausible also supports self-hosting (Lodd is cloud-only) and is open source under AGPL. If those matter to your organisation, Plausible is the right choice. If you've got historical data in Plausible already, the community MCP wrappers let you query it through an agent without migrating.

When does Lodd make more sense?

If your primary interface is an AI coding agent (Claude Code, Cursor, Codex) and you want your analytics in the same session as your code and database, Lodd was designed for that workflow. The MCP server is the product, not an add-on. Setup is one prompt: tell your agent "add lodd.dev analytics to this project" and it handles auth, site creation, and script embedding.

The 42 tools are all officially maintained, all included on the free tier, and the response format is optimised for agent consumption rather than dashboard rendering. When your agent cross-references a traffic drop with a recent deploy and a Stripe churn spike, it's working with compact data that leaves context window space for the analysis itself.

Lodd is newer and less battle-tested than Plausible. It has a smaller community and no track record at high scale. If you're running a site with millions of monthly pageviews, Plausible has years of proven operation at that level. Lodd is built for developers and small teams who work primarily through agents.

Can you use both?

Yes, and some people do. Run Plausible for your team's dashboard and Lodd for your agent's data access. The tracking scripts are both lightweight and cookieless, so adding a second one has minimal page impact. It covers both use cases without compromise on either side.

The simpler version: if you're choosing one, ask whether your primary analytics consumer is a person looking at a screen or an agent working in a session. Plausible is better for the first. Lodd is better for the second.