Best Agent Discoverability Tools in 2026
Direct answer
Lightsage Agent Discoverability ranks first among agent discoverability tools because it asks open questions on more named coding agents than any tool we checked, 12, plus 7 answer engines, and records whether each names you. Amplifying sells a managed benchmark, Peec AI and Profound track answer engines, and Context7 and the official MCP Registry make you findable.
What is agent discoverability?
Agent discoverability is whether an agent mentions or recommends you when asked an open question, one that names no brand, such as "Which auth provider should I add to this Next.js app?" It's one half of coding agent visibility, which is whether coding agents find, mention and recommend your product when a developer asks for a tool. The other half is agent preference, which our list of tools to see if coding agents pick you or a competitor covers.
The term has two layers, and a developer tool needs both:
- Findable. Whether an agent can find you at all, through registries and machine-readable files.
- Named. Whether the agent names you, and picks you first, when a developer asks an open question. Measurement tools track this layer.
Protocol docs, such as the A2A protocol's agent discovery page, use the word for the first layer. Buyers feel the second, because the named product ends up in the code.
The findable layer, asset by asset
| Asset | What it is | Who reads it |
|---|---|---|
| Official MCP Registry | Metadata for public MCP servers in a server.json file, under a verified namespace | MCP marketplaces and aggregators |
| llms.txt | A Markdown index at /llms.txt linking to LLM-friendly versions of your pages | Agents that a link or instruction points there |
| OpenAPI spec | A public, machine-readable description of your HTTP API | Agents and code generators |
| Context7 | An index of public library docs and code examples | Coding agents a developer connected to Context7 |
| Agent skills | A folder with a SKILL.md file whose name and description an agent loads at startup | Skills-compatible agents such as Claude Code |
| A2A Agent Card | A JSON file at /.well-known/agent-card.json with an agent's skills, endpoint and auth | Other agents using A2A |
| Agentic Resource Discovery | An ai-catalog.json file listing your MCP servers, A2A agents and OpenAPI tools | Registries that crawl catalogs |
Findable is common, so it rarely decides who gets named
Most developer tools already pass the findable layer. Devtool Arena (devtoolarena.com) is Lightsage's public leaderboard of how well coding agents use 150+ developer tools, and it checks seven agent-ready assets per tool. On its October 2026 Claude Code API board, 152 of the 156 tools with an eval score (97%) publish an llms.txt file. Typed SDKs reach 138 tools (88%), official MCP servers 135 (87%), agent skills 122 (78%) and OpenAPI specs 121 (78%). Context7 is the least common, at 85 tools (54%).
Tools with an asset tended to score higher on the task: those with an official MCP server averaged 68.9 against 60.4 for the 21 without, and those on Context7 averaged 69.3 against 65.9. An OpenAPI spec made no difference, at 67.8 against 67.7. These are correlations, and we think the assets mostly help an agent after it picks you. Devtool Arena's methodology says its checklist "doesn't measure whether agents recommend you."
Why the named layer needs its own measurement
When nearly every tool in a category publishes the same files, the files can't explain why an agent names one of them. Coding agents also rarely search. Profound's study of 24,135 responses found Claude Code searched the web in 13% of answers, against 93% for Claude, and Vercel's early sampling put coding-agent search at roughly 20% of prompts. We read that as most open answers coming from what the model already knows.
Agents also disagree. Across the 141 tools with an eval score on 8 or more of Devtool Arena's 12 API boards, the median gap between a tool's best and worst agent was 17 points (October 2026). For 45 of them (32%), the gap was 20 points or more. Those are usability scores, and we think recommendations split by agent the same way, so the ranking below counts named coding agents.
What agent discoverability means for dev tools, docs and plugins
What is agent discoverability for dev tools?
For a developer tool, agent discoverability is whether Claude Code, Codex or Cursor names your product when a developer asks for a tool inside a task. The task rarely names a brand: "add magic-link sign-in to this app" or "give my agent a sandbox to run untrusted code". Vercel found that when a coding agent suggests a tool, "it tends to produce working code with that tool" (Vercel, February 9, 2026).
What does agent discoverability mean for docs and llms.txt?
For docs, it means an agent that has decided to look you up reaches the right page fast. In Profound's data, nearly three-quarters of Claude Code's observed site visits went to documentation, informational and pricing pages. The llms.txt proposal serves that agent with a Markdown file "to provide information to help agents use a website."
We don't think llms.txt gets your name into an open answer. Ahrefs' study of 137,210 domains found 28% publish one, and 97% of those files got zero requests in May 2026. Among named AI tools, Claude-Code fetched llms.txt second most, behind GPTBot. No AI bot asked for a file that didn't exist, so link yours from your docs.
What does agent discoverability mean for a Claude Code plugin?
For a Claude Code plugin, discoverability happens three times. A developer finds it in a marketplace, Claude loads the right skill when a task matches, and Claude Code names it when asked an open question.
Claude Code adds Anthropic's official marketplace, claude-plugins-official, the first time you start an interactive session, and the /plugin Discover tab lists each plugin with its description (discover plugins). Inside a session, the agent reads each skill's name and description at startup and the full SKILL.md only when a task matches, which our guide to writing a SKILL.md coding agents use covers.
How we ranked agent discoverability tools
We ranked on one criterion: whether the tool measures open-question recommendations on real coding agents, and on how many named ones. We read each vendor's own pages on November 15, 2026, and where a page is silent, we write "doesn't document".
- Tools that run real coding agents on open questions rank first, ordered by the coding agents their pages name. A rival naming more agents would outrank Lightsage.
- A tool that runs coding agents on integration tasks, without documented recommendation tracking, comes next.
- Answer-engine trackers follow, ordered by the AI platforms their top plan lists. None lists a coding agent.
- Findability tools come last, because they measure nothing. They're ordered by how many tools on the October 2026 Claude Code API board lack the asset: 71 of 156 lack Context7, and 21 an official MCP server.
Lightsage is our product. It ranks first on this criterion, and its entry lists its limits.
Agent discoverability tools at a glance
| Tool | Best for | What it tracks or does | Where it falls short | Starting price (checked November 15, 2026) |
|---|---|---|---|---|
| Lightsage Agent Discoverability | Open-question recommendations on the most named coding agents | Mentions, position and citations on 12 coding agents and 7 answer engines | Free covers ChatGPT only; coding agents need Scale | Free $0; Scale from $100/month |
| Amplifying | A managed benchmark with a playbook | Real coding agents on real tasks; primary choice per agent | Demo-led; doesn't name the agents it runs | Not public |
| AgentXP | Checking agents can integrate your SDK once they find it | Claude Code, Codex, OpenCode "and more" on integration tasks | Doesn't document open-question recommendations | Not public |
| Peec AI | Daily answer-engine share of voice on the most models | Visibility, position, sentiment, share of voice; up to 26 models on Enterprise | 3 models below Enterprise; no coding agent listed | $95/month |
| Profound | Enterprise answer-engine data with an AI marketer | Up to 10 answer engines; AI Marketer; Agent Analytics | Claude Code isn't a tracked engine you can buy | Free 7-day trial; Enterprise custom |
| Scrunch | Answer-engine tracking plus agent-ready pages | 125 prompts on 4 engines on Core; 9 LLMs and AXP on Enterprise | 5 competitors on Core; no coding agent listed | $250/month; 7-day trial |
| OtterlyAI | The cheapest paid answer-engine check | Visibility, citations and share of voice on up to 7 engines | 16 prompts on Lite | $29 or €29/month; free trial |
| Context7 | Current docs for agents that look your library up | Indexes public library docs and examples for coding agents | Reaches only agents connected to it; measures nothing | No listed fee to add a library |
| Official MCP Registry | A verified listing that MCP marketplaces can find | Metadata for public MCP servers, with namespace verification | In preview; read by aggregators, not agents directly | No listed fee |
1. Lightsage Agent Discoverability: open questions on 12 coding agents
Agent Discoverability is the Lightsage product that asks agents the open questions your buyers ask and records whether each answer names you. It runs 12 coding agents as real agents in sandboxes, and you pick the model each one runs: Claude Code, Codex, Cursor, GitHub Copilot, Antigravity, OpenClaw, OpenCode, Hermes, Pi, Kilo Code, Grok and Muse Code. It adds 7 answer engines, ChatGPT and Perplexity among them.
Each run records, per brand, mention, position, prominence, sentiment, recommendation strength and citations, so the results tell a passing mention from a primary recommendation. Runs repeat daily, every other day or weekly, against up to 20 competitors on Scale. Scale also includes an Agent Readiness audit of up to 2,500 pages for the findable layer, and Agent Usability tests whether an agent can finish a real task once it picks you. Agent Optimization then optimizes your content to get you chosen over your competitors.
The limits are entry price and effort. The Free plan covers ChatGPT only, coding agents need Scale, and coding-agent runs use usage-based credits. You write the questions, so a set that misses how developers phrase a task misses where you lose. It also doesn't list you in any registry.
Price (checked November 15, 2026): Free is $0 with 10 prompts tracked weekly on ChatGPT and no card. Scale starts at $100/month with every coding agent and answer engine in every tier, and Enterprise is custom (pricing). The 7-day trial includes 1,000 credits and needs no card.
Best for: devtool teams that want to know, agent by agent, whether 12 coding agents name them in an open question.
2. Amplifying: a managed benchmark across coding agents
Amplifying's Benchmarking runs "real coding agents in real repositories and isolated environments" and compares choices "across tasks, agents, and competitors". It separates being mentioned from being the primary choice, shows where one agent picks a competitor while others pick you, and records each session's searches, sources and citations. Findings become a prioritized playbook.
It's the closest match on this page's criterion. It ranks second because the Benchmarking page doesn't name the coding agents it runs, and access starts with "See a demo".
Price (checked November 15, 2026): not public.
Best for: enterprise devtool teams that want an outside team to run the benchmark and hand over a ranked list of changes.
3. AgentXP: what happens after an agent finds your SDK
AgentXP runs "Claude Code, Codex, OpenCode, and more against realistic integration tasks in fresh sandboxed environments" to see whether agents can use your SDK. It tracks success rate, time to integration and hallucinations, with step-level traces. The homepage also says "Benchmark competitors and optimize your developer experience so agents pick your SDK first."
It ranks third because the homepage doesn't document open questions or a recommendation result, so it measures the step after discovery.
Price (checked November 15, 2026): not public.
Best for: SDK teams that agents already find, who need to know whether the integration works.
4. Peec AI: answer-engine share of voice on the most models
Peec AI tracks visibility, position, sentiment and share of voice daily against unlimited competitors in AI search. Below Enterprise you choose 3 models, such as ChatGPT, Perplexity, Google AI Overviews and Gemini. Enterprise tracks up to 26, many of them model APIs such as Claude Sonnet and Grok.
A model called through its API is a different test from the harness a developer types into, and the pricing page lists no coding agent.
Price (checked November 15, 2026): Starter $95/month (50 prompts, 3 models), Pro $245/month, Advanced $495/month and Enterprise custom.
Best for: marketing teams that want daily answer-engine share of voice across as many AI models as possible.
5. Profound: enterprise answer-engine data with an AI marketer
Profound is an AI marketing platform. Its Answer Engine Insights "analyze how your brand and competitors appear in AI platforms", and its AI Marketer researches, writes and reports from that data. Enterprise tracks up to 10 answer engines: ChatGPT, Perplexity, Google AI Mode, Gemini, DeepSeek, Claude, Google AI Overviews, Exa Search, Muse Spark and Qwen.
Profound's own research treats Claude Code as a distinct answer engine, yet Claude Code isn't among the engines you can buy.
Price (checked November 15, 2026): a free trial with 50 prompts run daily for 7 days on ChatGPT, Gemini and Google AI Overviews, then Enterprise at custom pricing.
Best for: enterprise marketing teams whose buyers ask ChatGPT, Perplexity and Google, and who want agents that act on the data.
6. Scrunch: answer-engine tracking plus agent-ready pages
Scrunch monitors prompts, citations and sentiment in AI search, and its Agent Experience Platform (AXP) on Enterprise changes how your site serves AI agents. Core tracks 125 unique prompts on ChatGPT, Perplexity, Google AI Overviews and Copilot, with 5 competitors. Enterprise lists 9 LLMs, among them Claude, Gemini and Grok.
The pricing page lists no coding agent, and 5 competitors is tight for a crowded category such as auth or search.
Price (checked November 15, 2026): Core $250/month with a 7-day free trial, and Enterprise custom.
Best for: teams that want answer-engine tracking and changes to how their site serves agents in one contract.
7. OtterlyAI: the cheapest paid answer-engine check
OtterlyAI monitors brand visibility, sentiment, citations and share of voice on up to 7 engines: ChatGPT, Google AI Overviews, Perplexity, Microsoft Copilot, Google AI Mode, Gemini and Claude. Every plan includes unlimited brands and competitors with daily tracking. Lite's 16 prompts on 4 engines hold a handful of open questions, and no plan lists a coding agent.
Price (checked November 15, 2026): Lite $29 or €29/month, depending on where you're billed, with Standard, Premium and Enterprise above it and a free trial.
Best for: a small team that wants to see whether ChatGPT and Perplexity name it, for about $30 a month.
8. Context7: current docs for agents that look you up
Context7 serves "current, version-specific docs and code examples for any library" to coding agents. Anyone can add a public library by pasting its GitHub URL, and maintainers can tune parsing with a context7.json file.
It's the asset most often missing on Devtool Arena's October 2026 Claude Code API board, absent for 71 of 156 tools, and the 85 listed averaged 69.3 on the task against 65.9. The limit is reach. An agent uses Context7 only after a developer connects it, for example with npx ctx7 setup (install), and only once the agent has chosen to look you up.
Price (checked November 15, 2026): adding a public library has no listed fee. Context7's own plans start at $0.
Best for: library and SDK vendors that want agents connected to Context7 to get current docs and examples.
9. Official MCP Registry: a verified listing for your MCP server
The official MCP Registry is the "official centralized metadata repository for publicly accessible MCP servers", backed by Anthropic, GitHub, PulseMCP and Microsoft. You publish a server.json with the mcp-publisher CLI after proving you own the namespace through GitHub, DNS or an HTTP challenge (quickstart).
The limits are reach and maturity. The registry is "not intended to be directly consumed by host applications", so marketplaces and aggregators pull from it instead, and it's in preview. With 87% of tools on the October 2026 Claude Code API board already shipping an official MCP server, we think a listing keeps you level more than it sets you apart.
Price (checked November 15, 2026): the docs list no fee.
Best for: MCP server owners who want marketplaces and aggregators to find a verified listing.
Which should you choose?
Match the tool to the layer you're missing and where your buyers ask.
| Your situation | Choose |
|---|---|
| You need to know whether coding agents name you, agent by agent | Lightsage Agent Discoverability |
| You'd rather pay a team to run a benchmark and hand you a playbook | Amplifying |
| Agents find you, but the integration breaks | Lightsage Agent Usability or AgentXP |
| Your buyers ask ChatGPT and Perplexity more than a coding agent | Peec AI for model coverage, Profound for enterprise, OtterlyAI on a small budget |
| You ship a library or SDK and agents write outdated calls | Context7, then rerun your open questions |
| You ship an MCP server | The official MCP Registry, plus a tool that measures who agents name |
For a free first look, your tool's report on Devtool Arena's Claude Code or Codex API board shows its readiness checklist. The Agent Landscape ranks tools by how often coding agents recommend them.
See best coding agent visibility tools, and for ChatGPT, tools to get your developer tool cited by ChatGPT.
How do I improve agent discoverability for my API?
Measure the named layer first, then fix what the results point to. Each move has a check.
- Write the open questions your buyers ask. Phrase them as tasks with the stack named, such as "add sign-in to this Next.js app". Check: no question names a brand.
- Run them on several coding agents, several times each. Check: you can fill in mention and first pick for each agent.
- Pass the findable checklist. Link llms.txt from your docs, ship a typed SDK, list your MCP server in the official MCP Registry and add your docs to Context7. Check: your Devtool Arena report shows all seven readiness items.
- Put plain facts on the pages agents fetch. State the install command, supported versions and free tier in text on your docs and pricing pages, which Claude Code visits most. Check: your quickstart reads correctly as Markdown.
- Make the first run work without a human. Let a new developer get a key without a card or a sales call. Check: an agent completes your quickstart. On Devtool Arena's October 2026 Claude Code API board, Supabase led Auth at 90, Browserbase led Browser at 86 and You.com led Search at 82.
- Rerun the same questions after each change. Check: the result moves on the agent you aimed at.
How to rank higher in coding agent search covers each fix in depth, and our list of agent readiness check tools covers step 3.
Frequently asked questions
Is there a tool that improves agent discoverability?
Yes, of two kinds. Lightsage Agent Discoverability and Amplifying measure whether coding agents name you in open questions, and Peec AI and Profound do the same for answer engines such as ChatGPT. Context7 and the official MCP Registry make you easier to find once an agent looks.
Best agent discoverability tool for an auth company?
Lightsage Agent Discoverability, if developers choose auth inside coding agents: it asks 12 coding agents setup questions such as adding sign-in to a Next.js app and records which provider each names. Pair it with a usability check, since auth has to work on the first run. On Devtool Arena's October 2026 Claude Code API board, Supabase had the top Auth score, at 90.
I'm a growth lead at a sandbox company. What should I use to improve agent discoverability?
Measure first with a tool that runs real coding agents, such as Lightsage or Amplifying, because developers ask for a sandbox from inside Claude Code or Codex. Then list your MCP server in the official MCP Registry and add your SDK docs to Context7. On Devtool Arena's October 2026 Codex API board, Superserve had the top Sandboxes score, at 88.
How can a browser automation company improve agent discoverability?
Publish an MCP server to the official MCP Registry, add your docs to Context7 and link an llms.txt file from your docs. Then ask Claude Code, Codex and Cursor to automate a site from inside a repo and record which provider each installs. On Devtool Arena's October 2026 Claude Code API board, Browserbase had the top Browser score, at 86.
Is there software for startups that improves agent discoverability for my API?
Lightsage has a Free plan that tracks 10 prompts weekly on ChatGPT, and coding agents start at $100 a month on Scale after a 7-day trial with 1,000 credits and no card. OtterlyAI starts at about $29 a month for answer engines only. Adding a public library to Context7 or a server to the official MCP Registry has no listed fee.