Best Tools for DevRel Teams to Improve How AI Agents Use Your Product

Direct answer

DevRel teams need one tool per job. Use Lightsage to check whether AI coding agents recommend you, audit readiness, test whether they finish your quickstart through your API, CLI, MCP server or SKILL.md, and see which docs they read. Use Cloudflare's scanner for a free first check and Mintlify for agent-readable docs.

---

How we chose these tools

A developer advocate now serves two readers: the developer, and the AI coding agent the developer asks. What AgentRel is covers the discipline, and our AgentRel platform list ranks platforms by how many of its four jobs they measure. This page is the toolkit for the people doing the work, ordered by the five jobs on a DevRel team's list.

• The axis. Within each job, tools come first when they test or measure what AI coding agents do, and later when they only describe what a site contains.
• Evidence. Each vendor's own pages, read on 1 November 2026. Where a page is silent, we write "doesn't document".
• Lightsage. We make it, so we rank it first only where a named criterion supports that, and we list a limitation for it in each job.
• Scope. The tools here are ones a DevRel team can start without engineering support. Tools that sit in your CI pipeline are covered in our MCP server testing list.

Which tool does which DevRel job?

Job 1: find out whether AI agents recommend you

To find out whether AI agents recommend you, ask them the questions your buyers ask and count the answers. Agent discoverability is whether an agent mentions or recommends you when asked an open question. Agent preference is whether it picks you over a named competitor.

Lightsage covers 19 platforms: 12 coding agents and 7 answer engines. Agent Discoverability runs your tracked prompts daily and records, per brand, whether you're mentioned, your position, prominence, sentiment, recommendation strength and citations. Agent Preference runs a head-to-head between you and one named competitor, with 1 to 20 prompts, up to 10 agents and 1 to 20 runs per prompt and agent. A judge records won, lost or tie, and you get a win rate. Our criterion for ranking it first is breadth of real coding agents run in sandboxes. The limitation is price of entry: the Free plan tracks 10 prompts a week on ChatGPT only, and coding agents start on Scale at $100/month (pricing, 1 November 2026).

Profound tracks answer engines. Its pricing page, read 1 November 2026, lists a free trial with 50 prompts daily for 7 days on ChatGPT, Gemini and Google AI Overviews, and an Enterprise plan with up to nine answer engines and custom pricing. It fits a brand team whose buyers search in ChatGPT. Its pricing page doesn't document coding agents such as Claude Code or Cursor as tracked engines.

Devtool Arena is a free public leaderboard at devtoolarena.com covering 150+ developer tools. Its Agent Landscape ranks tools by how often coding agents recommend them, so you can see where you sit before paying for anything.

Best for: Lightsage when coding agents are your buyers, Profound when answer engines are.

Job 2: run an agent readiness check on your docs and site

An agent readiness check scans your site for the files and signals AI agents use to find and read it. Our list of agent readiness check tools compares the options in more depth. Here, "agent readiness" means how readable your site is to AI agents, which is unrelated to a company's AI-adoption maturity.

Cloudflare's Is Your Site Agent-Ready? is free and checks several standards across five categories: discoverability, content accessibility, bot access control, protocol discovery and commerce. It covers robots.txt, Markdown content negotiation, MCP Server Cards, Agent Skills and OAuth discovery. Paste the fix prompt it gives you into your coding agent. It grades what exists on your site rather than what an agent does with it.

agent-ready.dev scores a site against the Vercel Agent Readability Spec (15 site-wide and 25 per-page checks), llms.txt (10 checks) and agent protocols (15 checks), and offers weekly monitoring. Its homepage doesn't state a price. Pick it when you want a score that doesn't come from a hosting vendor.

Lightsage's Agent Readiness Audit sits under Agent Discoverability and crawls every page of your website and docs, in separate Website and Documentation modes. For each page it checks AI crawler access, whether content shows without JavaScript, whether the page returns Markdown when an AI agent asks, page length, load time, metadata and whether pricing, docs, API reference and quickstart can be reached. It ranks fixes by severity and gives a copyable prompt for your coding agent per fix. It's included on Scale (up to 2,500 pages) and Enterprise, not Free. Its Benchmarks view also runs Cloudflare's scan inside the app and saves the results. Our criterion is page coverage: it audits the whole docs site, where the free scanners look at the pages you give them.

A passing score doesn't mean agents succeed. In Devtool Arena's Claude Code API board for October 2026, 46 of 155 scored tools had a perfect checklist score. They averaged 71.7 on the task, against 62.5 for the 31 tools under 70, and still ranged from 39 to 91. Microsoft's DevAX site adds that it hasn't found evidence agents read llms.txt. Treat these checks as hygiene, and move to Job 3 for proof.

Best for: a free first pass with Cloudflare, a whole-site crawl with Lightsage.

Job 3: test whether AI coding agents can finish your quickstart

To test a quickstart, give real coding agents the task a new developer would type, and record whether they finish. This is agent usability: whether an agent can complete a real task with your product.

Lightsage Agent Usability runs your task against your API, CLI, MCP server or a custom task in sandboxes. You get pass or fail, full traces, errors, and per-agent and per-endpoint scores. Skills Evals runs the same eval with and without your agent skill and compares them, which answers whether your SKILL.md helps. See Agent Usability and our guide to running agent evals. Our criterion is the number of agents per eval; the limitation is that it needs Scale.

Microsoft Scope is an MIT-licensed, research-preview platform for measuring how coding agents handle your product. You define a task and criteria, and it runs repeated experiments. GitHub Copilot and Claude Code are supported today, Codex and OpenCode are planned and Cursor isn't integrated (Scope site, 1 November 2026). It runs on Kubernetes, so a DevRel team will need engineering help to host it. Choose it when you want to own the infrastructure and the data.

Devtool Arena shows what a finished result looks like. Each company page lists the task prompt, tool calls, runtime and cost. On the Claude Code board, tools scoring 90 or above had a median of 6.5 tool calls and 0.5 errors per run, against 24 and 4.3 for tools scoring 60 to 79. On the Cursor board the medians were 11 and 0 against 40 and 2.6. More calls and more errors mean the agent is stuck, so use them as your debugging signal.

Best for: Lightsage, for a hosted run across many agents.

Job 4: see which docs pages AI agents read and which signups they drive

To see which docs pages agents read, you need to tell agent visits from human ones. Agent traffic is the visits to your site and docs that come from coding agents, browsing agents and bots.

Agent Analytics from Lightsage is middleware (@lightsage/agent-tracker) for Next.js, Vercel, Cloudflare, Express and Mintlify. It classifies visits as coding agent, browsing agent, bot or human, and attributes signups and revenue to agent visits. Its audit also shows agent visits and AI citations per page. Our criterion is signup attribution. The limitation is that it's middleware, so someone must install it on your docs host.

Profound's Agent Analytics tracks AI-sourced traffic and attribution across unlimited domains, with integrations listed for Akamai, AWS, Cloudflare, Fastly, Google Analytics, Google Cloud Platform, Netlify, Vercel and WordPress (pricing page, 1 November 2026). It suits a team that already buys Profound for answer engines.

Mintlify is a docs host rather than an analytics tool. When you host there, every page sends Markdown to AI agents, every page is available at a .md URL, and Mintlify hosts llms.txt and skill.md files and a docs MCP server. That changes what agents read. You'll still need an analytics tool to see who read it, and the Lightsage tracker supports Mintlify sites.

Start with the question "which pages do agents fetch before a signup" and compare it to your top human pages. We think the two lists differ.

Best for: Lightsage for visits tied to signups, Mintlify for changing what agents receive.

Job 5: make your CLI and MCP server easier for AI agents to use

To make your CLI and MCP server easier for AI agents to use, change one thing at a time and re-run the same task. Microsoft's DevAX site measured eight changes against a bare baseline, using GitHub Copilot on SharePoint Framework upgrade tasks. Four of them sit with a DevRel team:

Passing CLI arguments as one JSON payload cost 4 to 11 times more per task with no gain in correctness. DevAX also found that fixing the release notes the agent had already read made every run use the CLI for every developer, with no install. The lesson for DevRel is to edit the page the agent already reads, and to write the warning, not the tip. For the CLI itself, see how to make your CLI agent-friendly.

Context7 serves current, version-specific docs to Claude, Codex, Cursor and other agents (context7.com). In DevAX's test its MCP server was called in 0 of 5 runs: its tools didn't load in 3 and weren't called in the other 2. That's one task with one harness. In Devtool Arena's Claude Code API board, the 79 of 143 tools with a Context7 entry averaged 71.4 against 67.9 without (October 2026). Keep the listing, and measure whether your agents call it.

For your own server, Lightsage's CLI and MCP evals run the before-and-after test. Our MCP server testing list covers narrower tools. The DevAX site also ships an AX Practitioner skill, installed with npx skills add microsoft/scope/docs/ax-playbook/ax-practitioner, that answers playbook questions in your coding agent.

Best for: DevAX's playbook as the method, a Lightsage eval as the test.

Which should you choose?

1. A one-person DevRel team with no budget. Run Cloudflare's scanner and agent-ready.dev, read the DevAX playbook, and check where you rank on Devtool Arena. Then try the Lightsage 7-day trial, which includes 1,000 credits and every agent with no card.
2. A small team shipping an API or CLI. Use Lightsage Agent Usability for pass or fail on your quickstart, and Agent Discoverability for recommendations. Fix the docs page the agent read first.
3. A brand team whose buyers use ChatGPT. Start with Profound for answer engines, and add Lightsage when developers' coding agents begin to choose tools for them.
4. A docs-heavy product on a site you want to change fast. Move or mirror docs to Mintlify for Markdown and skill.md, then re-run your quickstart eval to check it helped.

Pricing for Lightsage is on the pricing page: Free is $0, Scale starts at $100/month with every agent included, and Enterprise is custom.

Frequently asked questions

Our DevRel team wants to optimize for AI coding agents. What tool should we use?

Pick a tool per job. Use Lightsage to measure whether AI coding agents recommend you and whether they finish your quickstart through your API, CLI or MCP server. Use Cloudflare's scanner or agent-ready.dev for a free readiness check, and Mintlify if you want agent-readable docs by default.

As a developer advocate, how do I measure how agents experience my product?

Give each AI coding agent the same real task, such as your quickstart, run it several times, and record pass or fail, tool calls, errors and cost. Change one thing, such as a docs warning or a CLI flag, and run the same task again. Lightsage Agent Usability automates this loop across 12 coding agents.

Our developer advocate team wants to improve how agents use our SKILL.md agent skill. What tool should we use?

Run the same task with and without the skill and compare the results. Lightsage Skills Evals does this across coding agents in sandboxes, and Microsoft's DevAX playbook describes the method and ships an AX Practitioner skill. Check that the agent loads the skill at all before you edit its text.

As a developer advocate, how do I track coding agent traffic?

Add middleware that classifies each visit as a coding agent, browsing agent, bot or human, and join it to signups. Lightsage Agent Analytics does this for Next.js, Vercel, Cloudflare, Express and Mintlify sites. Profound's Agent Analytics reads CDN and server logs and suits enterprise brand teams.

Is there a free way for a DevRel team to run an agent readiness check?

Yes. Cloudflare's Is Your Site Agent-Ready scanner and agent-ready.dev both offer a scan from their homepages, and Devtool Arena publishes a free readiness checklist per tool. They check files and standards. They don't show whether an AI coding agent can finish a task, which needs an eval.