> AI agents: this is one page of nextanalytica's site. The full index for agents: https://nextanalytica.app/llms.txt

# nextanalytica MCP server

For AI agents that act for a user. Built 2026-11-18 from the server's own tool list. The full index for agents is [llms.txt](https://nextanalytica.app/llms.txt).

nextanalytica is headless web analytics for your own AI agent: one tag collects your site's visits, your agent (Claude, Cursor, Codex or any MCP client) reads them through our MCP server, and skills on a schedule tell your team what changed and what to fix. Everything it offers: [platform](https://nextanalytica.app/platform.md). Prices: [pricing](https://nextanalytica.app/pricing.md).

## What to do for your user

- **They want nextanalytica connected to this client:** add the remote MCP server `https://mcp.nextanalytica.app/mcp`. In Claude Code: `claude mcp add --transport http nextanalytica https://mcp.nextanalytica.app/mcp`. In the Claude app: Settings, Connectors, Add custom connector. Ask before you run a command on their machine.
- **Signing in is their step:** the server uses OAuth in the browser (in Claude Code: `/mcp`, then nextanalytica). You cannot do it for them; tell them how in one line and wait.
- **They want their website measured:** call `list_projects`. If the site isn't there, they add it at https://dash.nextanalytica.app. Then call `get_tag` for the site, add the tag to its code, and show them the change.
- **They want an answer about their site:** call `get_context` for the project first, `describe_data` before asking for numbers, then `query`, `find_changes` or `get_dashboard`. Never invent a number; say what the data can't answer.
- **They want checks on a schedule:** `list_skills` shows the skills, `get_skill` loads one. Scheduled skills run on the Claude Code of a teammate who allowed it, never on our servers.
- **They ask what's new:** read [Changes](https://nextanalytica.app/changes.md) before saying a capability doesn't exist.

## Summary

- **What it is:** a hosted MCP server over a website's analytics: visits, pages, sources and channels (AI assistants are their own channel), conversions, funnels, retention and paths, connected ad, search and revenue sources, the project's memory, skills and connectors.
- **Endpoint:** `https://mcp.nextanalytica.app/mcp` (streamable HTTP). Nothing to host.
- **Auth:** OAuth per person, or a personal key made in the dashboard, sent as `Authorization: Bearer`. There are no shared keys.
- **Price:** connecting and calling tools is free, and 30 days of history are free. A longer history will be a paid option; nothing is charged without the user's yes. We run no model: thinking happens on the user's own agent.
- **Data:** read [Security](https://nextanalytica.app/security.md) for how projects are kept apart and what is stored.

## When it fits

- The user wants to know how their website is doing and would rather ask in chat than learn a dashboard.
- They want to see visits from AI assistants (ChatGPT, Perplexity, Claude and others) next to search, ads and the rest.
- They want Google Search Console, Google Analytics 4, Bing Webmaster Tools, ChatGPT Ads, Odoo numbers read next to the visits they came from.
- They want checks that run on a schedule and report in chat.

## When it does not fit

- The data lives in another analytics tool and the user won't add our tag. This server reads nextanalytica projects only (GA4 can be connected as a source).
- Mobile apps: the tag is for websites for now.
- A long history right away: a project keeps 30 days, and tools refuse a period past what is kept.
- Bulk exports of raw events: this is a read API for agents, not an export pipeline.
- Session replays, feature flags and A/B tests: not offered.

## Good to know

- Tools that change something (marked below) may need an owner or admin on the project; the tool says so when you aren't one.
- Repeated reads of the same question are served from a one-minute cache per project.
- A project's numbers are only seen by the people and agents its team allows.

## Server instructions

nextanalytica holds this person's web analytics. Start with get_context for the project, read describe_data before asking for numbers, then use query, find_changes and get_dashboard. Never invent a number; say what the data can't answer.

## Tools that read

- `list_projects`: The projects you have access to, with your role on each.
- `get_context`: Call first for a project, before every answer: its facts (region, time zone, currency, kept days, tag status), the layers in the order they win (our base rules, the team's instructions, your skills, your own instructions), the team's memory as information with who wrote each…
- `list_skills`: Your skills on this project (each with its schedule), the skills your teammates use, and built-in skills nobody has on yet. Names and descriptions only: get_skill gives one skill's instructions. Skills are switched on, written and copied in the Skills window.
- `get_skill`: One skill's instructions and your schedule for it. Load a skill before doing its job, then follow it within the project instructions, which win over it.
- `list_connectors`: The tools your team connected to this project (Linear, HubSpot, Wix, their own…) that you may act through now, with each one's actions and what they take. Blocked ones are left out. Act with call_connector, and only when the person asks or a skill says so.
- `describe_data`: What every metric and dimension means, which events and properties this project really sends, its conversions, source freshness and the kept window. Read it before query.
- `get_dashboard`: The basic dashboard in one call, blocks in this order: overview, pages, referrers and UTM, countries, devices, os, events, speed (with the Real Experience Score and routes in Poor, Needs improvement and Great), AI traffic and conversions, each against the period before.
- `query`: Ask for numbers with a spec, names from describe_data only.
- `seo_queries`: Google searches that showed the site: clicks, impressions, click-through rate and average position, from Search Console. min_position 8, max_position 15 finds the queries almost on page one; page narrows to one page.
- `seo_pages`: The site's pages in Google Search: clicks, impressions, click-through rate and average position, from Search Console. Order by impressions with a low ctr to find pages that rank but lose clicks.
- `find_changes`: What moved a metric against the period before: the biggest contributors by channel, campaign, landing page, page, device and country. Use it for 'why did X drop or grow'.
- `get_timeline`: Dated business events in a period: launches, campaigns, price changes, sales, outages, site and tag changes, notes. Check it when a number moved.
- `list_insights`: Signals from our rules and findings from your skills, newest first, with their numbers and the query that proves them.
- `get_live`: Who is on the site right now: visitors and page views in the last five minutes, their pages and channels.
- `get_person`: One person (a visitor who consented or a user the site identified): properties, ids, groups, visits, and their latest 100 events. Pass a user id, a distinct id or a person id.
- `find_people`: People who did something (a page, an event or a conversion), at least min_times, optionally without another step, optionally with a property, in a period, optionally only those in a saved cohort: how many, and up to 50 of them. Only consented or identified people.
- `get_tag`: The project's tag and how to add it: the script line, where it goes in common frameworks (Next.js, React, Vue, Nuxt, SvelteKit, Astro, plain HTML), whether each domain has been seen, and how consent and identify work.

## Tools that change something

- `get_skills_plugin`: For coding agents that take plugins (Claude Code): a private address that installs your skills on this project as a plugin, kept in step with the Skills window. Run the two commands it returns. Calling it again makes a new address and the old one stops working.
- `call_connector`: Act through one of the project's connectors: run one of its actions with arguments, as list_connectors describes them. Logged with your name; refused if the connector is blocked. Get the person's yes before anything that changes something.
- `suggest_ideas`: Leave up to 6 ideas on this person's project page, made from what the numbers showed in your conversations: tasks that leave something behind (knowledge, a report, a check, a change), never questions. Each is a card they copy back to you.
- `set_runner_consent`: Record the person's answer to: "May your scheduled skills run on your Claude? " Only the skills this person scheduled run there, never a teammate's.
- `start_run`: Start a scheduled run: with the ticket you were started with, or with no ticket (an hourly routine pulling its work). Returns the skill to run, whom it reports to, the period to look at, the tools this run may use and the run's ticket; with no ticket and nothing due, says so.
- `post_finding`: Report what this run found to the skill's owner: a short title and summary, the numbers, and the query spec that proves them. One finding per run; posting again updates it.
- `suggest_memory`: Leave a short note in the team's memory from this run: a business fact, a decision, or what worked, with its numbers.
- `finish_run`: End this run: done, or failed with the reason. Always the last call of a run.
- `set_conversions`: Say what counts as a conversion for this project (replaces the list). Rules: {event} a custom event, {form, step} a form submit, {path} a page (end with * for a prefix), {order: true}.
- `add_note`: Add a dated event to the project's timeline (a launch, a price change, a campaign, a site change), so later answers can explain a jump.
- `remember`: Save a short note to the team's memory after the person agrees: a business fact, a decision, or what worked with its numbers.
- `mark_insight`: Mark an insight seen, or dismissed (a dismissed subject stays quiet for you for 30 days).
- `link_campaign`: Tell us which utm_campaign an ad platform's campaign uses, when describe_data lists it under campaign_gaps, so its spend meets its conversions.
- `save_cohort`: Save a group of people to come back to, by the same rules as find_people (what they did, didn't, how often, a property, a period). Saving an existing key replaces it, when you saved it (or you are an owner or admin).
- `delete_cohort`: Delete a saved cohort: its definition and its list of people. The member who saved it, or an owner or admin, can. The people themselves stay.
- `forget_person`: Erase one of the site's visitors for good, when they ask for it (GDPR): the person and every id they had are gone from every answer at once, and from storage within 30 days. Pass the person id, user id or distinct id.

Canonical address: https://nextanalytica.app/mcp
