---
title: Agent Community
description: Documentation for Agent Community — what the community is, how it is governed, how to join, how to connect an agent, and every machine-readable surface this site publishes.
---

# The agentic web needs shared infrastructure

Agents are shipping fast. They handle support, schedule meetings, negotiate deals, and coordinate across systems. But the ecosystem is fragmented — no shared naming, no discovery standard, no community governance.

That's a problem. Without common infrastructure, one platform will own the layer between agents and the internet. When that happens, everyone else plays by their rules.

**We're building the alternative.** An open community of the people building the agentic web, working in public on agent identity, discovery, and trust.

---

## For agents

We argue that agents should be discoverable through open standards rather than hardcoded configuration. It would be strange to argue that and not do it, so this domain implements the standards it advocates. Everything below is live today on ordinary domains, needs no API key, no registration and no approval from us, and is one command away from being checked.

| I want to… | Use | Path | Reference |
|---|---|---|---|
| **Resolve a domain to its agent** | AID, one DNS TXT lookup | `_agent.<domain>` TXT | [AID spec](/docs/aid) |
| **Call this community's tools** | MCP, Streamable HTTP, JSON-RPC over POST | `/mcp` | [MCP server](/docs/mcp-server) |
| **Read the tool schemas** | MCP manifest | [`/.well-known/mcp.json`](https://agentcommunity.org/.well-known/mcp.json) | [MCP server](/docs/mcp-server) |
| **Same capabilities over A2A** | A2A agent card | [`/.well-known/agent-card.json`](https://agentcommunity.org/.well-known/agent-card.json) | [MCP server](/docs/mcp-server) |
| **Read stats or content** | REST, GET, no key | `/api/v1/...` | [API reference](/docs/api-reference) |
| **Machine-read the API** | OpenAPI 3.1 | [`/openapi.json`](https://agentcommunity.org/openapi.json) | [API reference](/docs/api-reference) |
| **Receive events** | Webhooks, HMAC-SHA256 | you host the receiver | [Webhooks](/docs/webhooks) |
| **Search these docs** | A second MCP server | `/mcp/docs` | [Documentation MCP](/docs/docs-mcp) |
| **Ask in natural language** | NLWeb v0.55, GET or POST | `/ask` | [NLWeb](/docs/nlweb) |
| **Install a published skill** | Agent skills | [`/.well-known/agent-skills/index.json`](https://agentcommunity.org/.well-known/agent-skills/index.json) | — |
| **Read any page as clean text** | append `.md` | `/docs/aid.md`, `/faq.md` | — |
| **Ingest the whole site at once** | llms.txt | [`/llms.txt`](https://agentcommunity.org/llms.txt) | — |
| **Ingest just these docs** | docs-llms.txt | [`/docs-llms.txt`](https://agentcommunity.org/docs-llms.txt) | — |
| **Ingest just the API** | api-llms.txt | [`/api-llms.txt`](https://agentcommunity.org/api-llms.txt) | — |
| **Know what you may do here** | agents.md | [`/agents.md`](https://agentcommunity.org/agents.md) | — |
| **See who runs the community** | team.md | [`/team.md`](https://agentcommunity.org/team.md) | — |
| **Find the structured-data feeds** | Schema feed map | [`/schemamap.xml`](https://agentcommunity.org/schemamap.xml) | — |

**The four MCP tools, in full:** `lookup_member`, `get_community_stats`, `register_agent`, `verify_certificate`. That is the entire tool surface, deliberately — see [the MCP reference](/docs/mcp-server) for schemas. `register_agent` requires explicit authorization from the person you are acting for in the current turn.

**What `/mcp` answers.** Send JSON-RPC over `POST`; per the Streamable HTTP transport that is the whole protocol. `GET` with `Accept: text/event-stream` opens an SSE stream, for clients that expect one around `initialize`. A `GET` from a browser gets `405` and a pointer to the manifest.

If something in that table does not do what it says, that is a bug worth reporting. The whole point of publishing a list like this is that anyone can hold it to account, and a surface that quietly rots is worse than one that was never advertised.

### Start with one DNS lookup

This is [AID](/docs/aid), the discovery specification the community maintains, resolving this domain to its own agent endpoint:

```bash
dig +short TXT _agent.agentcommunity.org
```

```text
v=aid2;u=https://agentcommunity.org/mcp;p=mcp;a=none;s=Agent Community: member lookup, stats, .agent DMV;d=https://agentcommunity.org/docs/mcp-server
```

One lookup returns the endpoint, the protocol it speaks, the credentials it expects, and where its documentation lives. No registry, no sign-up, and no SDK needed to read it. The DMV publishes its own record the same way.

The reason to care about that record is what it removes. Without it, every integration hardcodes a URL in a config file somewhere, and every one of those breaks the day the endpoint moves. Email solved this in 1986 with the MX record; agents have no equivalent, so each platform invents its own. AID is the boring, unglamorous fix — a lookup any DNS client can already perform, with no central registry to ask permission from and no company positioned to charge rent on the answer.

Fastest path from nothing to a working call: [onboarding for agents](/docs/agent-onboarding).

---

## Start here

If you have never heard of any of this, the one-pager is the honest summary — what the community is, what has actually happened so far, and what is still just a proposal. "Why agents need names" is the argument underneath all of it. And if you already know why you are here and want to be counted, onboarding walks through registration, domain claims, and the endorsement letter.

<Cards>
  <Card
    href="/docs/one-pager"
    title="The one-pager"
    description="What the community is, what it is applying for, and where it stands — in about five minutes."
  />
  <Card
    href="/docs/why-agent"
    title="Why agents need names"
    description="Why agents need names, not just identifiers — and the case for a community-governed namespace."
  />
  <Card
    href="/docs/onboarding"
    title="Onboarding for members"
    description="Join as an individual or an organization, claim the domains you operate, and sign the endorsement letter."
  />
</Cards>

## Community

A community application is judged on whether the community it claims to represent actually exists, was organized before the application, and has a genuine connection to the string it wants. That is an unusually high bar for transparency, and it shapes what we publish: who is involved, how decisions are made, and what members have signed — all in public, all checkable by someone who is trying to disprove it.

These pages are that record. Governance covers how policy gets proposed and ratified. The team page names everyone, team and advisors alike, with their affiliations. The map is the public directory of member organizations and explains how a profile comes to exist. Agent Brief is our own reading of what is happening across the agentic web, which you can subscribe to without joining anything.

<Cards>
  <Card
    href="/docs/governance"
    title="Governance"
    description="How decisions get made. Working groups, community ballots, and policy published before it takes effect."
  />
  <Card
    href="/team"
    title="Team and advisors"
    description="Everyone behind the community, named, with affiliations — the team and the advisors."
  />
  <Card
    href="/docs/map"
    title="The map"
    description="How an organization gets a public profile: sign up, verify, we enrich it, then you edit it by hand or via an agent."
  />
  <Card
    href="/docs/newsletter"
    title="Agent Brief"
    description="Our curated snapshot of what is happening across the agentic web, daily or weekly."
  />
  <Card
    href="/docs/get-involved"
    title="Get involved"
    description="Working-group mailing lists, events, Discord, and the open specification repositories."
  />
  <Card
    href="/docs/work-items"
    title="Work items"
    description="What the working groups are building, and where the open problems are."
  />
</Cards>

## Technical

None of this waits on ICANN. The specifications and endpoints below work on ordinary domains today, and they would keep working whether or not a new top-level domain ever exists — that is deliberate, because infrastructure that only pays off after an approval is not infrastructure, it is a bet.

AID is the core of it: a discovery record that turns a domain name into an agent endpoint, with client libraries for TypeScript, Python, Go, Rust, Java, and .NET. The MCP and A2A surfaces are how you talk to this community's own data — member lookup, aggregate stats, pre-registration, and certificate verification, and nothing more, because a narrow tool surface is easier to reason about than a broad one. Webhooks push events out; the documentation MCP and NLWeb answers make these pages queryable rather than just readable.

<Cards>
  <Card
    href="/docs/agent-onboarding"
    title="Onboarding for agents"
    description="The fastest path from nothing to a working call: MCP, A2A, or the public REST API. No key required."
  />
  <Card
    href="/docs/aid"
    title="AID — Agent Identity & Discovery"
    description="Open specification for DNS-based discovery. One TXT record resolves a domain to an agent endpoint."
  />
  <Card
    href="/docs/mcp-server"
    title="MCP server"
    description="Four tools: member lookup, community stats, .agent pre-registration, and certificate verification."
  />
  <Card
    href="/docs/api-reference"
    title="API reference"
    description="The public REST reads — aggregate stats and published content, no key required."
  />
  <Card
    href="/docs/webhooks"
    title="Webhooks"
    description="Subscribe to community events. HMAC-SHA256 signed, with the verification recipe."
  />
  <Card
    href="/docs/docs-mcp"
    title="Documentation MCP"
    description="A second MCP server that searches and serves these docs directly to an agent."
  />
  <Card
    href="/docs/nlweb"
    title="NLWeb answers"
    description="Natural-language queries answered from the documentation corpus."
  />
</Cards>

---

## The gap right now

It is worth being precise about what is actually missing, because "agents need infrastructure" is vague enough to mean anything. Four specific things do not exist yet, and each one is a place where a default is being set right now by whoever moves first. None of them are hard technical problems. They are coordination problems, which is why they need a community rather than a clever protocol.

| Problem | What it means |
|---------|--------------|
| **No naming layer** | Agents are anonymous IDs. No brand, no accountability, no trust signal. |
| **No discovery standard** | Every platform invents its own way to find agents. Fragmentation slows adoption. |
| **No community governance** | Rules about agent identity are being set by platforms, not by builders. |
| **No shared trust** | Users can't tell a legitimate agent from a malicious one. |

## Where things live

| | |
|---|---|
| **[agentcommunity.org](https://agentcommunity.org)** | Community hub — standards, governance, membership |
| **[aid.agentcommunity.org](https://aid.agentcommunity.org)** | AID specification, workbench, and client libraries |
| **[dmv.agentcommunity.org](https://dmv.agentcommunity.org)** | Department of Machine Verification — agent identity pre-registration |
| **[The map](/map)** | Public directory of member organizations |
| **[news.agentcommunity.org](https://news.agentcommunity.org)** | Agent Brief — a curated snapshot of the agentic web |
| **[Working-group lists](/lists)** | Public mailing-list archives |
| **[GitHub](https://github.com/agentcommunity)** | Open source — specifications, SDKs, tooling |
| **[Discord](https://discord.gg/juPSbFBWZg)** | Real-time discussion |
| **[Blog](/blog)** | Updates and longer arguments |

---

*The agentic web is coming. The question is whether its infrastructure is open or owned. [Join the community](https://agentcommunity.org/join) to keep it open.*