What Is treg? The OpenRouter for AI Agent Tools, Explained

By Ammar Ahamed, Founder, Cocoon · 3 November 2026 · Facts checked against treg.to, its docs and its GitHub repository on 3 November 2026

What treg is

An AI agent like Claude Code can read your files, write code and run commands. What it can't do on its own is know things that live behind someone else's paywall: how many people search for a keyword, who links to a competitor, whether a company just raised money, or what is trending on TikTok today. That data sits behind subscriptions, signup walls and, in some cases, no public API at all.

treg is the layer that fixes that. You give your agent one treg token. The agent searches treg's catalog by the job it needs done ("backlinks for a domain", "find a work email", "keyword search volume"), reads the price, and calls the tool. treg injects the right provider's credential on its own server, relays the request, and hands back the provider's answer unchanged. The agent never holds a vendor key, and you never open a vendor account.

The comparison its makers use is OpenRouter. OpenRouter gives you one key for many language models. treg gives you one key for the tools and data an agent needs to finish real work. The line on the README sums up the idea: ask for the task, not the tool.

treg at a glance

Why treg exists: the seat problem

Say you want Claude Code to write a weekly SEO report. It needs keyword volumes, backlink counts and live search results. Doing that the usual way means a Semrush seat, a Moz seat and a SERP API plan, each billed monthly, for a report that makes a few hundred calls a week. treg's homepage adds up a typical SEO toolbelt at around $536 a month in seats.

Other data has no easy door at all. treg lists TikTok trends as invite-only, LinkedIn posts as partner-only, and Instagram data as gated behind app review. treg carries those accounts and routes, and bills you a fraction of a cent per call. Here is how its own homepage compares seats with per-call prices:

Source: treg.to homepage, 3 November 2026. Seat prices are the ones treg quotes.

The point isn't that seats are bad. If your team lives in the Semrush dashboard all day, keep it. The point is that an agent run is a different shape of work: many small lookups, across many vendors, in bursts. Paying per answer fits that shape better than paying per seat.

How treg works: search, read, call

Every job follows the same three steps, whether your agent uses the CLI or the MCP tools (catalog_search, catalog_get and call):

# 1. find tools that do the job, with prices
treg catalog search "backlinks for a domain"
# 2. read parameters, price and measured reliability
treg catalog get <id>
# 3. call it (or --data '{...}' for a JSON body)
treg call <id> --query key=value

The middle step is what makes treg more than a key vault. catalog get shows the exact price and, where several providers do the same job, compares them on measured success rate, speed and the last time each one answered. Nineteen providers sell backlink data and nine sell email lookups, so your agent picks on evidence instead of habit.

For common jobs, treg also offers routed tools named after the job rather than a vendor: treg.web.search, treg.web.extract, treg.google.serp.organic, treg.companies.enrich, treg.people.email.find. You send what you know, treg tries providers cheapest first, moves on when one misses, and on per-success providers bills only the answer it returns. When we enriched a company from its domain, the router skipped eight providers that needed inputs we didn't have (a LinkedIn URL, a name, an email) and answered from the first one that could.

Behind every catalog call sits a simple order of credentials, which treg calls the credential ladder:

1. Your team registered its own tool for that provider: that tool and that key.
2. Your team stored a key for the provider: that key.
3. Neither, and a verified free public route exists: no key, no charge.
4. Otherwise: treg's own account, billed to your team's balance.

If you already pay for a provider, connect your key once and those calls route through it, unmetered.

What's in the treg catalog

The catalog is grouped by what the endpoints do, not by vendor. On 3 November 2026 the main groups were:

• Keyword and rank tracking: keyword volume and ideas, competitor keywords, keyword difficulty, live Google results, rank tracking, plus Search Console and GA4 through your own OAuth connection.
• Backlinks and authority: backlinks and anchors, trust and citation flow, referring domains, broken-link checks.
• AI visibility: Google AI Overview citations, brand mentions in AI answers, cited-source tracking.
• Trending and discovery: TikTok trends and sounds, X posts and profiles, Instagram posts and reels, YouTube stats, subreddit posts, LinkedIn posts and pages.
• Publishing: post to X, LinkedIn and Instagram, upload to YouTube, through your own accounts.
• People and company enrichment: work email find and verify, profiles and role history, contact search, funding and investors, firmographics.
• Buying signals: hiring and headcount, job changes, local business data, deliverability checks.
• Ads: manage Google, Meta, TikTok and Microsoft Ads campaigns through OAuth, and research competitor creative in the public ad libraries.
• Measurement: GA4 conversions, Business Profile, Pinterest and Snapchat Ads, Slack messages, YouTube channel analytics, Google Trends.
• Generation: image, video and voice models, including Seedance, Veo, Wan, Gemini Image, GPT Image and Seedream, plus Fish Audio speech and voice cloning.

When the catalog has nothing for a job, the agent can file a request (treg catalog request "what you need"), and treg says those requests steer what it adds next.

How to set up treg in Claude Code, Codex or Cursor

The fastest way is to hand the job to your agent. Paste this one line into Claude Code, Codex or Cursor:

set up treg - https://treg.to/llms.txt

The agent reads treg's setup guide and runs the installer. If you'd rather do it yourself, it's three commands:

curl -fsSL https://treg.to/install.sh | sh
# browser sign-in: GitHub, Google or an emailed code
treg login
# shows your account, team, balance and next commands
treg

The installer adds the CLI and installs the treg skill for every agent it finds on your machine. In Claude Code you can also install it as a plugin:

/plugin marketplace add superdesigndev/treg
/plugin install treg@treg

On claude.ai or ChatGPT, there's nothing to install. Add https://treg.to/mcp/ as a custom connector and finish the OAuth consent, where you choose which team pays. treg also runs a separate surface for Claude's Connectors Directory at /mcp/v2/, which exposes curated catalog endpoints only, without image, video or audio generation, and keeps read calls apart from write calls.

If you've never wired an agent to an API before, our free API Lab has you send real API requests to a live feed, then hand the same API to an agent over MCP. It makes the rest of this guide easier to follow.

treg pricing: what a call actually costs

• Prepaid, per call. You top up a team balance; each call on treg's account is deducted at the listed price. The price is shown before the call.
• 0% markup. treg says it bills at the provider's own rate and earns on volume pricing with vendors, not on you. A few subscription-only providers carry a per-call rate treg set, and catalog get tells you whose price it is.
• $1.00 free once per new verified account, when it creates its first eligible team.
• Failed calls are free. A 4xx or 5xx error relayed from the provider costs nothing.
• Repeat questions are cheaper. When your team asks something it has asked before and treg serves the archived answer, that answer costs 10% of the price.
• Your own key is free. Calls that use a key your team registered are never metered.
• Out of balance returns an HTTP 402 with the estimated cost and a top-up link, so an agent can stop cleanly instead of failing silently.

Here is what our own first calls cost, run through treg on 3 November 2026 while researching this series:

Ten cents bought the keyword research behind all three posts in this series. The same research through seats would have needed at least two monthly subscriptions. We walk through the SEO calls in detail in Claude Code for SEO.

Your team's own keys, CLIs and skills

treg has a second half that matters just as much for teams: it's a safe home for the keys you already have. You can register an API key, an OAuth connection, a vendor CLI such as stripe or gh, or a SKILL.md recipe. Every teammate's agent can then call it, while the credential stays encrypted on the server.

# read-only preview of the keys, skills and CLIs it would register
treg scan
# register them (run again with --replace to update)
treg upload

Calling one of your own APIs is as simple as putting treg's prefix in front of the real URL. treg finds the tool by host, injects the key, strips your treg token, and relays everything else as it was:

GET https://treg.to/call/https://api.stripe.com/v1/charges
X-Treg-Token: <your token>

For a team, this solves three everyday problems:

• No keys in .env files on laptops. When someone leaves, you revoke one token instead of rotating ten vendor keys.
• Agents get their own identity. An admin can mint an agent token with treg org agent-new and limit it to certain tools, a spend cap and a pinned provider. Its calls show up under its own name.
• An audit trail. treg calls and treg audit show who called what, with results and costs, and treg health --run re-checks every stored credential.

Roles are simple: a viewer calls tools, a member also registers their own, an admin manages everyone's and invites, and an owner changes roles.

treg skills: ready-made workflows for your agent

A skill is a written recipe your agent follows end to end. treg ships several, each running its data steps through treg:

# add one skill
npx skills add superdesigndev/treg --skill lead-signals
# or the installer's whole set
treg skill bootstrap --skills all

We cover lead-signals and recruit in GTM Engineering With Claude Code.

treg vs a single-vendor MCP server, an MCP registry and a unified API

"MCP server" now covers very different things, so it helps to be precise. MCP (Model Context Protocol) is the standard way an agent connects to outside tools. Here is where treg sits among the options people search for:

These aren't exclusive. A common setup is the vendor's own MCP server for the one tool your team lives in, and treg for everything else. And if you connect your Semrush or Moz key to treg, your agent calls those through treg at no extra cost.

Open source and self-hosting

treg is public on GitHub at superdesigndev/treg, with more than 4,100 commits by early November 2026. The licence is Apache 2.0 with additional terms: free for internal and commercial use and for self-hosting, but not for offering it to others as a hosted service without written permission. You can run your own registry with your own provider keys, or list your own API or skill in the public catalog.

If you're building a product on treg that bills its own users, treg encourages it: tag each call with your customer's ID from your backend, set per-customer budgets, and invoice from treg's ledger.

What to watch out for

• Agents can spend faster than people. A loop over 5,000 rows at a cent each is $50. Give agents their own token with a spend cap, and ask the agent to show the expected total before any loop. treg's own guide tells agents to do exactly this.
• A found email or phone is a best match, not a fact. Verify before anyone sends to it. treg flags when a found email's domain doesn't match the company you asked about.
• Personal data is still personal data. Enrichment and people search touch real people's details. If you work with data about people in Sri Lanka, the PDPA's main duties apply from 1 January 2027; in Europe and the UK, GDPR applies now. Our PDPA guide covers what that means at work.
• Providers disagree. Two keyword tools can report different volumes for the same term. For decisions that matter, check a second provider.
• Answers come back in each provider's own format. treg relays and doesn't normalise (the routed tools add a small common layer), so field names differ between providers.
• A "free" route can occasionally bill if treg's account for that provider runs out and the call is relayed through another account. catalog get shows the overflow price when one applies.

Who treg is for, and who it isn't for

It fits marketers and growth teams who already work in Claude Code or Codex and want live SEO, social and ads data in the same place; GTM engineers building lead lists, enrichment and signals; founders who need answers from ten tools but can't justify ten seats; and developers putting data tools inside their own agents or products.

It's not for you if nobody on your team uses an AI agent yet, or if what you need is a visual dashboard to click through. In that case, a normal subscription to one good tool is simpler. Our best AI SEO tools roundup and the AI Tools Directory are better starting points.

The part a tool can't do for you

treg makes an agent capable. It doesn't make it careful. The people getting real value from this setup can do three things: brief an agent clearly, put a number on what it may spend and send, and check its output before it reaches a customer. Those are learnable skills, and they're the same whether the data comes from treg or anywhere else. If you want to see where agents fit in your own week first, map one process in our free Workflow Mapper, then try building it in Automation Lab.

Frequently asked questions

Sources

treg

• treg.to homepage
• treg setup guide (llms.txt)
• The treg skill: operating guide
• Building treg into your product
• superdesigndev/treg on GitHub (README, licence, skills)