Best Agent Analytics Tools for Developer Tools in 2026

Direct answer

Agent analytics tracks the AI agents that visit your product, such as coding agents reading your docs, and the signups that follow. For developer tools, Lightsage names each coding agent and ties AI-sourced visits to signups, Mintlify fits Mintlify-hosted docs, and Known Agents offers free agent counts. To monitor an agent you build, use Langfuse, LangSmith or Braintrust.

---

What is agent analytics?

Agent analytics is the measurement of AI agents that visit and use your product: which agents arrive, which pages they read, where they hit errors, and which signups and revenue follow. For a developer tool, the agents that matter most are coding agents such as Claude Code, Codex and Cursor, which read your docs while a developer builds with your API.

The same two words also describe LLM observability, which is monitoring agents you build yourself. Langfuse, LangSmith, Braintrust, Arize Phoenix and Datadog Agent Observability trace your own agent's prompts, tool calls, cost and latency, according to their pages as of November 2026. A third meaning is coding-agent usage inside your company: Amazon CloudWatch Coding Agent Insights charts your developers' token use and cost for Claude Code, Codex and GitHub Copilot.

A founder, DevRel lead or growth marketer at a devtool company wants the first meaning, and this page ranks tools for it. An AI engineer shipping an agent wants observability, and a platform team paying for coding-agent seats wants usage telemetry.

What a coding agent visit looks like

A coding agent visit is a short burst of plain HTTP requests to your docs, usually with no cookie and no analytics script. Devtool Arena (devtoolarena.com), Lightsage's public leaderboard of how well 12 coding agents use 150+ developer tools, gives each agent one developer task and makes it find the right docs, SDK and endpoints itself (methodology).

The visits that end in a working integration are quick. On the Claude Code API board (October 2026 batch), the 17 tools that scored 86 or more averaged 8.2 tool calls and 69 seconds per run, with 6 to 10 calls each. Stripe topped the board at 93 (A), with 6 tool calls in 36 seconds. We think this is why session-based analytics misses coding agents: the whole visit can be over in about a minute, and your JavaScript usually never runs. Our post on coding agent traffic that GA4 doesn't capture covers the mechanics.

Naming the agent is the second problem. Known Agents gives Claude Code's example user agent as Claude-User (claude-code/2.1.316; +https://support.anthropic.com/), which starts with the same token as Claude's chat fetcher, and notes that Claude Code publishes no verification method (Known Agents). We think a classifier that matches only the first token would file it under chat assistants.

Why coding agents need their own row in your analytics

Coding agents barely register across the whole web but make up close to half of the docs traffic Mintlify measures, so a general bot report buries them. As of 8 November 2026, Known Agents measured Claude Code at 0.0241% of all website traffic, Codex at 0.0010% and Cursor at 0.0007%, across the thousands of sites that use its Agent Analytics (Codex, Cursor).

Docs sites look nothing like that. Mintlify's March 2026 analysis of about 790 million documentation requests put AI coding agents at 45.3%, almost level with browsers at 45.8%. By October 2026, agents of all kinds made up 78% of traffic across Mintlify-hosted docs (Mintlify).

Coding agents also read different pages. In Profound's study of the 1,000 pages most visited by each agent from 18 July to 18 August 2026, nearly three-quarters of Claude Code's visits went to documentation, informational and pricing pages, against 5% for Claude's agent (Profound). A coding-agent row tells you a developer's agent read your quickstart and pricing page, which is a step in a purchase.

Counting crawlers and AI referrals for any website is a different job, and our list of AI traffic analytics tools covers it.

How we ranked agent analytics tools

The order comes from one test: how directly each tool identifies coding agents in live traffic, based on its own pages as of 8 November 2026. Ties go to the tool that also connects those visits to signups or revenue, because that's the next question a devtool team asks. We checked three things:

1. Coding agents as their own class. The tool names Claude Code, Codex, Cursor and others separately from crawlers and chat assistants. This sets the order.
2. Pages and errors per agent. You can see which docs pages each agent read and which ones returned errors.
3. Visits tied to outcomes. The tool connects AI-sourced visits to signups, API keys or revenue. This breaks ties.

Tools that don't document a coding-agent class follow, ordered by how close they get: PostHog takes custom bot rules, Profound separates real AI crawlers from spoofed ones, and Cloudflare's bot list stops at crawlers and assistants. Where a page is silent, we write "doesn't document", which isn't the same as "doesn't have". Microsoft Scope runs lab tests instead of reading traffic, so it's listed last for comparison.

Best agent analytics tools at a glance

1. Lightsage Agent Analytics: which coding agents read your docs, and which visits turn into signups

Lightsage Agent Analytics shows which AI agents visit your docs and site, what they read, and which of those visits lead to signups and revenue. It counts coding agents such as Claude Code, Codex, Cursor and GitHub Copilot as their own traffic type, apart from AI crawlers, AI search, assistant fetches, other bots and people. It works with Next.js, Vercel, Cloudflare, Express, Netlify and Mintlify.

For each coding agent, you see the pages it read and the pages that returned errors to it. On the outcome side, Lightsage connects visits from AI sources to signups, demo requests, checkouts and API keys. The same account runs Agent Discoverability, which tracks whether agents recommend you before they ever visit.

It ranks first because it treats coding agents as their own class and is the one tool on this list whose pages document tying AI-sourced visits to signups. Known Agents, in second, connects to more hosts.

The limitation is what counts as AI-sourced. Lightsage credits a signup to AI when the visitor arrives from an AI source, such as a link in a ChatGPT answer. When Claude Code reads your docs in a terminal and the developer later types your URL, that signup isn't credited to the agent.

Price (checked November 2026): Agent Analytics is a Scale feature, with 180 days of history and conversion tracking. Scale starts at $100/month with every coding agent and answer engine in every tier, Free covers 10 ChatGPT prompts tracked weekly, and Enterprise is custom (pricing). A 7-day trial includes 1,000 credits and needs no card.

Best for: devtool teams that want to see which coding agents read their docs and which AI-sourced visits turn into signups.

2. Known Agents: coding agents and API calls on any host, free to start

Known Agents is an analytics and control platform that shows every crawler, scraper and AI agent visiting your site in real time. Its directory covers 2,000+ bots across 15+ categories, including an AI Coding Agent category for agents that fetch "documentation and other resources to help build software". As of November 2026 that category lists 9 agents, including Claude Code, Codex, Cursor, GitHub Copilot, OpenCode and Devin (directory).

It ranks second because it names coding agents as their own class but doesn't document what happens after the visit. It connects to 16 platforms, more than any other tool on this list, from Vercel, Netlify and Cloudflare to AWS, Google Cloud and WordPress (pageview docs). You can slice traffic by agent, category and page, see which agents hit errors or blocked pages, and check whether a bot using a trusted name is spoofed. It also tracks REST API and MCP calls through a Node.js SDK or its API (API call docs).

The limitation is the last layer. The pages we read don't document tying a visit to a signup, an API key or a payment, so you'd join that data yourself.

Price (checked November 2026): Free is $0 for 1 project and 100,000 events a month, Startup is $29/month for 1 million events, Growth is $299/month for 10 million, and Enterprise is custom (pricing).

Best for: tracking coding-agent visits alongside the API and MCP calls agents make, on any host, from a free plan.

3. Mintlify: agent analytics inside hosted docs

Mintlify is a documentation platform that, since February 2026, shows which AI agents visited your docs, which pages they read most and what they searched for through MCP. It identifies agents by IP address and user agent while it serves the page, so there's no tracker to add (Mintlify guide). The mint analytics CLI splits human and agent traffic per page.

The guide also gives a baseline: across 25 companies over 7 days, the median docs site got 14 visits to llms.txt and 79 to llms-full.txt.

The limitations come from scope. Mintlify sees docs it hosts, so your marketing site and signup flow sit outside it, and its pages don't document signup attribution. Its guide also says Mintlify's analysis "could not separate Codex traffic from generic HTTP clients".

Price (checked November 2026): agent analytics is on Pro, at $450/month billed annually or $540/month billed monthly, and on Enterprise. The free Starter plan doesn't include it, and new accounts get a 14-day Pro trial (pricing).

Best for: docs teams already on Mintlify that want agent traffic with no setup.

4. PostHog: custom coding-agent rules beside your funnels

PostHog classifies every event by user agent and source IP as Regular, AI Agent, Bot or Automation, with AI crawler, AI search and AI assistant categories under AI Agent. Most bots don't run JavaScript, so you forward server or CDN logs as $http_log events through a Vercel logs source, an edge worker or the capture API. A Bots tab in Web Analytics then shows bot requests over time, each crawler and the most crawled paths (guide).

The strength is where the data lands. The traffic-type properties work as filters and breakdowns in trends, funnels, retention and paths, beside the signup and activation events you already send to PostHog.

The limitation is the categories. PostHog doesn't document a coding-agent class, so Claude Code and Codex need custom bot rules (up to 50 per project) filed under an existing category. The Bots tab is also in alpha as of November 2026.

Price (checked November 2026): the free tier includes 1 million analytics events a month with no card, and usage-based pricing applies after that (pricing).

Best for: product teams on PostHog that will write their own coding-agent rules and break them down in trends, retention and paths.

5. Profound Agent Analytics: whether AI crawls turn into human visits

Profound Agent Analytics shows which AI systems crawl your site, which pages they prefer and whether that activity turns into human visits. It reads server logs instead of a JavaScript tag, separates real AI crawlers from spoofed ones, and connects through Akamai, AWS, Cloudflare, Fastly, Google Analytics, Google Cloud Platform, Netlify, Vercel and WordPress (pricing). Its docs list OpenAI, Anthropic, Google, Perplexity, Microsoft, Apple, Meta and DeepSeek among the platforms it tracks (docs).

Its strength is the pairing with Profound's answer-engine tracking, so a marketing team sees crawls, citations and AI search referrals under one contract. The August 2026 study cited above shows Profound can see Claude Code's agent on the domains it tracks.

The limitation for a devtool team is the traffic model. The product pages describe AI crawlers and AI search, and they don't document coding agents as a traffic type you can filter, or signups tied to an agent visit.

Price (checked November 2026): Profound lists a free trial and custom Enterprise plans, and its pricing page shows Agent Analytics with unlimited domains tracked.

Best for: answer-engine teams that want to see which pages AI crawlers prefer and whether those crawls bring human visitors.

6. Cloudflare AI Crawl Control: Markdown checks and crawler data on Cloudflare

AI Crawl Control is Cloudflare's dashboard for monitoring and controlling AI crawlers, and it works on every Cloudflare plan with no setup. It shows requests by crawler, category, operator and host, status codes, the most requested paths, and referrals from AI platforms on paid plans (Analyze AI traffic).

One chart matters for docs teams. Content Format compares the content types AI systems ask for, by Accept header, with what your origin serves. In April 2026, Cloudflare reported that of 7 agents it tested in February, only Claude Code, OpenCode and Cursor asked for text/markdown by default (Cloudflare). It's worth checking that you serve Markdown when it's requested.

The limitation is coverage. The bot reference lists crawlers and assistants such as GPTBot, ClaudeBot and ChatGPT-User, and it doesn't document coding agents such as Claude Code or Codex, or any signup attribution.

Price (checked November 2026): included on all Cloudflare plans, including Free. Referral metrics need a paid plan.

Best for: docs teams on Cloudflare that want to check whether AI systems get Markdown when they ask for it.

7. Microsoft Scope: lab tests instead of live traffic

Microsoft Scope is an open-source platform that gives coding agents the same task, checks the result against criteria you define and repeats the run. As of November 2026 its site lists GitHub Copilot and Claude Code as supported, with OpenAI Codex and OpenCode planned.

Scope shows what an agent does with your product when you set the task, and it never sees who visits your docs. You host it on Kubernetes and write the tasks and criteria yourself.

Price (checked November 2026): free and open source. You run the infrastructure.

How do I track coding agent traffic?

Track coding agent traffic from the server, because coding agents fetch your pages over HTTP without running your analytics snippet. That's why AI agent traffic looks invisible or Direct in GA4. The first three steps cost nothing beyond the host you already pay for.

1. Read requests, not pageviews. Use CDN, hosting or middleware logs. On Vercel, the AI bots managed ruleset runs on every plan in Log mode (Vercel). On Cloudflare, AI Crawl Control works automatically.
2. Match coding agents by user agent. Search for strings such as claude-code and Cursor/, using Known Agents' directory for examples. Treat a match as a clue, since these agents publish no verification method.
3. Keep docs routes only. Count docs pages, Markdown endpoints, llms.txt and API specs, and drop images, fonts and scripts, as Mintlify's guide recommends.
4. Join visits to outcomes. Send an event when a developer signs up or creates an API key, with the AI source attached. Lightsage does this for you, and with other tools you build the join in your own analytics.

Our guide to seeing which docs pages coding agents read covers steps 2 and 3 in more depth.

Which should you choose?

Start from who is asking and which layer of the funnel you're missing.

The vertical rarely changes the pick: an inference, sandbox, text-to-speech, auth, email or payments API asks which coding agents read the docs and whether a developer signed up afterwards. As an illustration, a payments API like Stripe would watch which page returns a 404 to Codex.

If agents also call your MCP server, see best MCP analytics tools. Lightsage isn't worth paying for until coding agents read your docs, and a content site that needs only crawler counts is covered by the free plans from Cloudflare or Known Agents. To see whether agents recommend you before they ever visit, start with coding agent visibility tools.

Frequently asked questions

Is there a platform that does agent analytics across Codex and Claude Code?

Yes. As of November 2026, Lightsage Agent Analytics and Known Agents both list Claude Code and Codex among the coding agents they detect, and Lightsage also tracks the signups and API keys that follow AI-sourced visits. Mintlify reports agent traffic on docs it hosts, but its own guide says it can't separate Codex from generic HTTP clients.

Is there a platform in 2026 that can track coding agent traffic across Grok and Muse Code?

Not by name, as of November 2026. Neither Lightsage nor Known Agents lists Grok or Muse Code among the coding agents it detects. Lightsage does list every coding agent and bot that visited your site in a date range, so check that list or your raw logs for their user agents, and treat a self-reported user agent as a clue rather than proof.

What's the cheapest tool for agent analytics for an auth company?

Cloudflare AI Crawl Control is free on every Cloudflare plan, PostHog includes 1 million analytics events a month for free, and Known Agents' Free plan covers one site and 100,000 events a month. None of the three documents which AI-sourced visitors go on to sign up, so an auth company that needs that answer pays for Lightsage Scale, from $100 a month.

Is there software that does agent analytics for my docs and llms.txt?

Yes. Mintlify reports agent visits per page on docs it hosts, and Lightsage reports the pages each coding agent reads on docs you host yourself. Agents fetch llms.txt over plain HTTP like any other page, so a request-level tool records it as a path.