What MatteringAgent is and who runs it
Mattering Agents is an app where the people in a workspace set up AI agents that act on their behalf. The built-in agent is called Matter. When one of these agents needs a page from the public web, the request is made by MatteringAgent.
MatteringAgent is operated by Matt Block, Inc., doing business as Mattering. It is the same software for every customer, and it identifies itself the same way for every customer.
Every request starts with a person
MatteringAgent does not decide on its own to visit a site. Each request happens because a person asked their agent to do something, or because a person set up a routine that runs on their behalf, and that task needs a particular page.
In the language bot directories use, Mattering is an intermediary: an agent service that many customers use, where a different person starts each task. Mattering runs the software and is accountable for how it behaves. The person using it decides which pages their task needs.
The agents reach the web in two ways:
- The page reader. A plain HTTPS client that fetches one page at a time and extracts its text, so the agent can read the page and cite it.
- The agent browser. A headless Chrome browser the agent drives for interactive tasks a person asks for, such as opening a page, filling in a form, or using a site the customer has an account on. The browser is switched on by the workspace, one agent at a time.
How to recognise it
User agent strings
The page reader sends exactly this User-Agent header:
MatteringAgent/1.0 (+https://mattering.io/bots)
The agent browser keeps the standard Chrome User-Agent string, so sites render normally, and appends the same token to the end of it:
<standard Chrome User-Agent string> MatteringAgent/1.0 (+https://mattering.io/bots)
Both strings contain MatteringAgent, so a single rule that matches that token covers both.
Signed requests (Web Bot Auth)
A user agent string can be copied by anyone, so it is a label, not proof. The page reader also signs its
requests with Web Bot Auth: HTTP Message Signatures (RFC 9421) using the tag web-bot-auth.
Each signed request names where its keys live in a Signature-Agent header:
Signature-Agent: "https://agents.mattering.io"
The public keys you need to check those signatures are published in our key directory:
https://agents.mattering.io/.well-known/http-message-signatures-directory
A request that verifies against a key in that directory came from Mattering. A request that claims to be MatteringAgent but carries no valid signature should be treated as unverified.
Where the traffic comes from
MatteringAgent runs on Google Cloud in the us-central1 region. We do not publish a fixed list of IP addresses for it, so an IP allowlist is not a reliable way to identify it. That is why we verify by signature rather than by address.
What it respects
Before and while fetching, the page reader follows the rules your site publishes:
-
robots.txt. It checks your robots.txt before fetching a page and does not fetch
anything disallowed for
MatteringAgentor for*. - Crawl-delay. It honours a Crawl-delay you set.
- Retry-After. When your server answers with a Retry-After header, it waits as asked.
- Request rates. It spaces out its requests to any one host and caps the number of requests in a single run.
What it never does
- It does not solve CAPTCHAs or try to get past any other human check. When one appears, the agent stops and asks its person to take over, or to supply the source another way.
- It does not hide what it is. There is no stealth mode, no fingerprint spoofing, and no rotating or residential proxies.
- It is not a crawler. It does not follow links to crawl a site, does not collect pages in bulk, and does not gather pages to train AI models. A fetch happens because a person's task needs that page.
- It does not show saved sign-in details to the AI model. When a customer saves a login for the agent browser to use on a site they have an account with, those details are never shown to the model.
How to allow, limit or block it
You do not need to do anything to allow MatteringAgent. To limit or block the page reader, add rules for
the MatteringAgent user agent to your robots.txt.
Block it from the whole site:
User-agent: MatteringAgent
Disallow: /
Keep it out of part of the site:
User-agent: MatteringAgent
Disallow: /account/
Disallow: /checkout/
Ask it to wait between requests, in seconds:
User-agent: MatteringAgent
Crawl-delay: 10
Because the page reader also obeys the rules in your User-agent: * group, a path you disallow
for every user agent stays off-limits to MatteringAgent even if you add a MatteringAgent group.
You can also block or rate-limit it at your firewall or CDN. Matching the MatteringAgent token
in the User-Agent header covers both the page reader and the agent browser. Matching a verified Web Bot
Auth signature is the more reliable rule for the page reader, because it cannot be copied.
How to report a problem
If MatteringAgent is behaving in a way you do not expect on your site, tell us through our contact form. Choose Support and start your message with MatteringAgent. It helps to include your site's address, the approximate time of the requests with a time zone, and the User-Agent or signature details you saw.
Mattering customers who want to know how this affects their own agents can read How your agents access the web.