Best Coding Agent Visibility Tools in 2026

Direct answer

Lightsage ranks first for coding agent visibility because it tracks recommendations on more named coding agents than any tool we checked in October 2026: 12, plus 7 answer engines, with preference, usability evals and agent traffic. Agent Discoverability suits MCP servers, Amplifying sells benchmarks, and Known Agents and Mintlify show which agents read your site.

---

What should a coding agent visibility tool measure?

A coding agent visibility tool should show which product each coding agent picks when a developer asks for one, agent by agent. Coding agent visibility is whether coding agents such as Claude Code, Codex and Cursor find, mention and recommend your product. We score every tool below against four questions:

• Recommend: agent discoverability is whether an agent mentions or recommends you when asked an open question, such as "Which API should my Python agent use to search the web?"
• Prefer: agent preference is whether an agent picks you over a named competitor.
• Use: agent usability is whether an agent can complete a real task with your product, from sign-up to a working integration.
• Visit: agent traffic is which coding agents read your site and docs.

What separates a useful tracker from a noisy one

Five design choices decide whether a tracker's numbers hold up. We read each tool's docs against them.

1. It runs the real agent. A coding agent answers inside a harness with its own system prompt, tools and search rules, so querying the model's API is a different test. Profound's August 2026 study of 24,135 responses found Claude Code searched the web in 13% of answers against 93% for Claude, and the brands the two mentioned overlapped by only about 20%.
2. It reads the code as well as the prose. Vercel wrote in February 2026 that when a coding agent suggests a tool, "it tends to produce working code with that tool." The same post found coding agents searched the web in roughly 20% of prompts, which suggests most picks come from what the model already knows.
3. It separates mentioned from picked. An agent can list five options without choosing one, write its own code instead, or mention a product while advising against it. We think the pick matters more than the mention, because the picked product is the one that ends up in the code. Recording position and prominence per brand keeps "named third" and "installed" from scoring the same.
4. It keeps failed runs out of the score. Runs error, time out or come back empty, and counting them as "not mentioned" drops your score whenever a provider has a bad hour. One run per prompt is also an anecdote. In the same Profound study, two Claude Code responses to the same prompt shared only about 2 in 5 of the brands they mentioned. We read that as a reason to run each prompt several times before trusting a change.
5. It tests usability separately. A recommendation says nothing about whether the agent can then sign up, authenticate and ship a working integration. Devtool Arena, our public usability leaderboard, shows the spread: on its Claude Code API board (October 2026), four web search APIs scored 74 to 81 on the same task and grader, with Firecrawl at 81, TinyFish 79, Tavily 77 and Exa 74.

How we ranked the tools

We included tools that run or observe a coding agent to measure what agents recommend, pick, use or read.

• The axis: the number of coding agents a tool tracks recommendations on, as named on its own pages on October 19 and 20, 2026.
• Ties: the tool that answers more of the other three questions ranks higher.
• No recommendation tracking: these tools come last, ordered by the coding agents they name.
• Evidence: each vendor's own pages, checked October 19 and 20, 2026. Where a page is silent, we write "doesn't document".

We left out chat-focused trackers such as Peec AI, Otterly.AI and AthenaHQ, which list Claude Code only as an MCP client.

The seven tools at a glance

1. Lightsage: recommendations, preference, usability and traffic on 12 coding agents

Lightsage is a self-serve platform that measures how coding agents and answer engines discover, choose and use developer tools. It covers 19 platforms: 12 coding agents and 7 answer engines. The coding agents are Claude Code, Codex, Cursor, GitHub Copilot, Antigravity, OpenClaw, OpenCode, Hermes, Pi, Kilo Code, Grok and Muse Code, each run as a real agent in a sandbox.

It answers all four questions in one account:

• Agent Discoverability runs tracked prompts daily and records each brand's mention, position, prominence, sentiment, recommendation strength and citations. Runs that fail at the provider stay out of the visibility score.
• Agent Preference puts you against one tracked competitor on up to 10 agents. A judge records won, lost or tie for each run, and the result is a win rate.
• Agent Usability has real coding agents run your task against your API, MCP server or CLI in sandboxes, with pass or fail and full traces.
• Agent Analytics is middleware (@lightsage/agent-tracker) that classifies visits as coding agent, browsing agent, bot or human, and attributes signups and revenue to agent visits.

Devtool Arena is the public version of these usability tests, covering 150+ developer tools. It publishes October 2026 API boards for Claude Code, Codex, Muse Code, Gemini and Grok. Devtool Arena's Agent Landscape ranks tools by how often coding agents recommend them.

Its main limitation is plan coverage. The $100 Startup plan includes only Claude Code and Codex among coding agents. Cursor, GitHub Copilot, OpenCode and four more need Growth at $1,000/month, and Antigravity, OpenClaw and Hermes need Enterprise. Coding-agent runs use usage-based credits, so a large prompt set across many agents uses them faster. Agent Preference has no landing page yet.

Price (checked October 2026): Startup $100/month, Growth $1,000/month, Enterprise custom (pricing). The 7-day trial includes 1,000 credits and every coding agent, with no card.

Best for: devtool teams that want all four answers on the most named coding agents, in one self-serve account.

2. Agent Discoverability: built for MCP servers and connectors

Agent Discoverability is a separate company from Lightsage's product of the same name. It helps companies with an MCP server, ChatGPT app or Claude connector get surfaced and picked. Its FAQ says: "Run discovery tests in ChatGPT, Claude and Cowork, plus the Claude Code and Codex coding harnesses." It also offers a free MCP Discoverability Score and agents that draft MCP description changes.

That focus is its limitation. It names two coding harnesses and doesn't document SDK, API or CLI adoption or usability evals, so a company selling an API would need a second tool.

Price (checked October 2026): not public; the score check is free.

Best for: companies whose main agent surface is an MCP server, ChatGPT app or Claude connector.

3. Amplifying: a benchmark with a playbook

Amplifying runs "real coding agents in real repositories and isolated environments" and compares choices "across tasks, agents, and competitors" (benchmarking). It separates being mentioned from being the primary choice, records each session's searches and sources, and turns findings into a prioritized playbook for positioning, docs and tooling. Ground Control reruns a task before and after a change.

Its benchmarking page describes four views: a leaderboard of primary choices, alternatives and mentions; an agent comparison that shows where one agent picks a competitor; recorded agent journeys; and the domains and pages agents cite. Each finding links back to the sessions and citations behind it.

It's aimed at enterprise devtool and infrastructure teams. The engagement is demo-only, and its benchmarking page doesn't name which coding agents it runs.

Price (checked October 2026): not public ("See a demo").

Best for: enterprise teams that want a research-style benchmark turned into a ranked list of changes.

4. Known Agents: coding-agent visits on any website

Known Agents tracks the crawlers, scrapers and AI agents that visit your website in real time, and adds AI chat referral tracking, an automatic robots.txt and an agent identification API. Its agent directory lists nine AI coding agents that fetch "documentation and other resources to help build software," among them Claude Code, Codex, Cursor, GitHub Copilot and OpenCode. Its Claude-Code page reported that 7% of top websites block Claude-Code in robots.txt, as of October 19, 2026.

It suits teams whose docs sit outside Mintlify or GitBook, or that want agent visits across the marketing site too. Like every traffic tool, it sees an agent only after the agent reaches you, so a developer whose agent picked a competitor never appears. The same Claude-Code page warns that because Claude Code "does not publish a verification method, any client can claim its identity," so a user-agent match is only a clue.

Price (checked October 20, 2026): Free for one project and 100K events a month, Startup $29/month, Growth $299/month, Enterprise custom (pricing). Its pricing FAQ suggests budgeting about twice your monthly human visits in events.

Best for: seeing which coding agents visit any website you run, starting on a free plan.

5. Mintlify agent analytics: which agents read your docs

Mintlify's agent analytics identifies Claude Code and Cursor doc fetches by user agent and IP on Mintlify-hosted docs, along with pages read and MCP searches. Its state of agent traffic report found Claude Code made 25.2% and Cursor 18.0% of all requests to Mintlify docs. Its own guide says it "could not separate Codex traffic from generic HTTP clients."

It sees only agents that already reached your docs, so the developer whose agent recommended a competitor never shows up. GitBook's site analytics has a similar Agents & MCP report for GitBook-hosted docs, covering the coding agents and crawlers that read a site through Markdown, llms.txt and its MCP server.

Price (checked October 2026): Starter is free; agent analytics comes with Pro at $450/month billed annually ($540 monthly) and Enterprise (pricing).

Best for: seeing which docs pages Claude Code and Cursor read, on docs you already host with Mintlify.

6. Profound: Claude Code research, answer engines in the product

Profound is an AI marketing platform for enterprise teams, built on answer-engine data across nine engines on its Enterprise plan. Its August 24, 2026 research treated Claude and Claude Code as distinct answer engines, and its Agent Analytics separates Claude Code's fetches from Claude's.

Claude Code isn't on its list of tracked engines you can buy (checked October 19, 2026), and its site doesn't document usability tests for APIs, CLIs or MCP servers. Our page on Profound alternatives that track coding agents compares it in more depth.

Price (checked October 2026): a free 7-day trial (50 prompts, three engines, no API) and Enterprise at custom pricing (pricing).

Best for: enterprise marketing teams whose main concern is ChatGPT, Perplexity and Google, with coding agents a research question for now.

7. Ahrefs Bot Analytics: coding-agent crawls next to your SEO data

Ahrefs Bot Analytics records bot traffic server-side through Cloudflare, Vercel, Amazon CloudFront or Fastly, with no JavaScript, and sorts it into 12+ categories such as AI Crawler, AI Assistant and AI Search. Ahrefs used it for a June 2026 study of 137,210 domains that classified requests by individual user agent. In May 2026, 97% of the llms.txt files those domains published got no traffic at all. Among named AI tools, Claude-Code fetched llms.txt second most often, behind GPTBot and ahead of every AI search and assistant bot.

It fits SEO and content teams that already pay for Ahrefs. Its product page doesn't document a coding-agent category among the 12+, so you pick out Claude-Code by user agent, as the study did. Like the other traffic tools, it can't show what an agent recommended, and it's still in beta.

Price (checked October 20, 2026): free during the beta on all paid Ahrefs plans, which start with Lite at $129/month (pricing).

Best for: SEO teams already on Ahrefs that want coding-agent crawls beside their search-engine crawl data.

Which coding agent visibility tool should you choose?

Skip Lightsage if coding agents never touch your product, or if ChatGPT and Perplexity are your only concern. Profound fits that job better.

With no budget, run a prompt by hand several times in each agent and log each run's winner. As an illustration, a search company like Tavily could run claude -p "Which API should my Python agent use to search the web?" and the same prompt through codex exec. How to see if coding agents pick a competitor covers head-to-head checks, and how to rank higher in coding agent search covers the fixes.

Frequently asked questions

Is there a tool that tracks coding agent visibility?

Yes. As of October 2026, Lightsage, Amplifying and Agent Discoverability measure what coding agents such as Claude Code and Codex recommend or pick. Known Agents, Mintlify, Profound and Ahrefs Bot Analytics observe coding agents that visit your site or docs.

Our growth team wants to track coding agent visibility. What's the cheapest tool we should use?

Lightsage's Startup plan, at $100 a month as of October 2026, is the lowest published price we found among tools that track coding-agent recommendations. It covers Claude Code and Codex, with a 7-day trial and no card. Agent Discoverability and Amplifying don't publish prices, and Known Agents has a free plan if you only need to see which coding agents visit your site.

Is there a platform that does coding agent visibility across OpenCode and Cursor?

Lightsage tracks recommendations on both on its Growth plan and above. Known Agents lists OpenCode and Cursor among the coding agents it identifies when they visit your site, and Mintlify's agent analytics identifies Cursor doc fetches on Mintlify-hosted docs.

Is there software that helps my MCP server get picked by coding agents?

Agent Discoverability is built for that: it runs discovery tests for MCP servers and connectors in Claude Code, Codex, ChatGPT and Claude, and drafts description changes. To check that agents can then use the server, Lightsage tests MCP servers alongside APIs and CLIs in sandboxes.

What's the best enterprise tool to track coding agent visibility for a developer platform company?

Choose software or a service first. For software, Lightsage Enterprise covers all 12 of its coding agents, while Profound Enterprise covers nine answer engines but doesn't sell Claude Code as a tracked engine. For a service, Amplifying sells benchmarks with a prioritized playbook.