--- title: "Agentic Atlas developer resources · Agentic Atlas" description: "Every developer resource Agentic Atlas publishes, with its address: the hosted MCP server, the read-only JSON API, the OpenAPI description, and authentication." canonical: "https://agentic-atlas.dev/developers" last-updated: "2026-09-23" --- For developers and agents # Agentic Atlas developer resources Every surface this domain publishes for you to build on, with its address. Nothing here needs a key, an account, or an install. ## Start here Agentic Atlas publishes one sealed Release of a curated corpus of agentic design patterns, over two transports: a hosted [MCP server](https://agentic-atlas.dev/connect) for an agent mid-task, and a read-only [JSON API](https://agentic-atlas.dev/api-reference) for an application. Both read the same Release, and every answer read from it names it: the JSON API in `X-Catalog-Revision`, the MCP transport in each payload's coherence marker, so a caller can always tell whether two answers describe the same Atlas. This page is the index. The contracts live at the addresses below, and each one is authoritative for itself. If you are an agent reading rather than a person, [/llms.txt](https://agentic-atlas.dev/llms.txt) is this same list in the manifest dialect, and [/agents.md](https://agentic-atlas.dev/agents.md) and [the AI catalog](https://agentic-atlas.dev/.well-known/ai-catalog.json) are its machine spellings. ## Agentic Atlas MCP server Stateless Streamable HTTP, hosted, unauthenticated, and the surface to prefer if you are wiring up an agent rather than an application. There is nothing to install and no local corpus to cache: the endpoint serves the one active Release, and every payload names it so two answers can be told apart. - [Agentic Atlas MCP endpoint](https://agentic-atlas.dev/mcp/) (MCP) — The transport itself. JSON-RPC over POST; a GET answers with a connection descriptor rather than a bare 405. - [Connect an agent to Agentic Atlas](https://agentic-atlas.dev/connect) (HTML) — The two ways in, the tool grammar, and a worked first call. - [Agentic Atlas consultation contract](https://agentic-atlas.dev/consult.md) (Markdown) — The rules that govern every call, written to be copied into a system prompt verbatim. - [Agentic Atlas MCP server card](https://agentic-atlas.dev/.well-known/mcp-server-card) (JSON) — The registry card: name, version, transport. - [Agentic Atlas official MCP registry listing](https://registry.modelcontextprotocol.io/v0.1/servers?search=dev.agentic-atlas/atlas) (Registry) — The verified domain-scoped registry row for the hosted transport. - [Agentic Atlas on Glama](https://glama.ai/mcp/connectors/dev.agentic-atlas/atlas) (Registry) — The connector row, keyed to the same domain-scoped name and claimed from this origin. - [Agentic Atlas in the Claude Connectors Directory](https://claude.ai/directory/agentic-atlas) (Registry) — The publisher's community listing, screened by Anthropic but not domain-verified. It connects Claude straight to the endpoint above, with no sign-in. - [Agentic Atlas on MCP.so](https://mcp.so/servers/agentic-atlas-bf5d87) (Registry) — The directory row, linking back to this origin and its endpoint. - [Agentic Atlas on Smithery](https://smithery.ai/servers/aj604/agentic-atlas) (Registry) — The publisher's account-scoped row. It reaches this server through Smithery's gateway, which needs a Smithery key; the endpoint above needs none. - [Agentic Atlas MCP tool card](https://agentic-atlas.dev/.well-known/mcp/server-card.json) (JSON) — The endpoint and the complete frozen tool grammar. - [Agentic Atlas plugin for Claude Code](https://github.com/aj604/agentic-atlas-plugin) (Source) — The public companion repository: three skills over two read-only subagents, on the hosted transport it registers for you. - [Agentic Atlas Agent Skills index](https://agentic-atlas.dev/.well-known/agent-skills/index.json) (JSON) — The same three skills over HTTP, in the Agent Skills discovery format: each one's SKILL.md address and a SHA-256 digest of it. No install, no registry, no connection. - [Agentic Atlas skill: reviewing-agent-designs](https://www.skills.sh/aj604/agentic-atlas-plugin/reviewing-agent-designs) (skills.sh) — Audit an existing agent artifact against the published patterns. - [Agentic Atlas skill: designing-agent-systems](https://www.skills.sh/aj604/agentic-atlas-plugin/designing-agent-systems) (skills.sh) — Design a new agent system with cited Agentic Atlas patterns. - [Agentic Atlas skill: applying-a-pattern](https://www.skills.sh/aj604/agentic-atlas-plugin/applying-a-pattern) (skills.sh) — Turn one named pattern into a cited edit plan for an artifact. ## Agentic Atlas HTTP API Read-only JSON under /api/. No key, no account, no OAuth, and no write operations of any kind. Reads are bounded by an explicit limit rather than paged, and 300 requests per minute per address is the published ceiling — successful and refused responses state the window so a client can throttle itself, except the direct 400 for an unsupported API-Version, whose sender carries no limiter state. - [Agentic Atlas API reference](https://agentic-atlas.dev/api-reference) (HTML) — Every operation in prose, with its parameters, bounds and refusals. Server-rendered, so it reads without running JavaScript. - [Agentic Atlas OpenAPI description](https://agentic-atlas.dev/openapi.json) (OpenAPI) — The authoritative document, and the one to generate a client from. - [Agentic Atlas browsable API reference](https://agentic-atlas.dev/docs) (HTML) — The same description in Swagger UI. Needs a browser that runs scripts. - [Agentic Atlas pattern index](https://agentic-atlas.dev/api/tree) (JSON) — The canonical traversal of every published pattern. - [Agentic Atlas batch read](https://agentic-atlas.dev/api/cards) (JSON) — `?ids=a&ids=b` reads a Card for each id in one call. Atomic: one unadmitted id fails the request. - [Agentic Atlas natural-language search](https://agentic-atlas.dev/ask) (JSON) — The NLWeb endpoint. POST a question for ranked records with their addresses; retrieval only, nothing is generated. Limited to the API's numbers in a window of its own, which `RateLimit-Policy` names. - [Agentic Atlas A2A agent card](https://agentic-atlas.dev/.well-known/agent-card.json) (JSON) — A2A 1.0 JSON-RPC at `/a2a`. Send `A2A-Version: 1.0` and a `SendMessage` request with a `ROLE_USER` message, a `messageId`, and `parts: [{"text": "your query"}]`. Returns the Agent Orient envelope in a direct message; no generated advice or retained conversation. Optional `metadata.expected_revision` pins the Release. Authentication, limits, and refusals are in `/auth.md`. ## Agentic Atlas authentication There is none, and that is the answer rather than a gap in this page. No account, no API key, no token, no OAuth, no authorization server. Read the first of these before you build an authentication step against this domain — including the part that says a `WWW-Authenticate` header from this origin means you are not talking to this deployment. - [Agentic Atlas authentication document](https://agentic-atlas.dev/auth.md) (Markdown) — The auth.md convention's own step order, with every step answered for a resource that issues no credentials. ## Agentic Atlas availability What this deployment does when it fails, and where the live answer is. There is no service level agreement, no uptime history and no third-party status page, and the first address here says so rather than leaving it to be inferred — along with every refusal you can receive and the recovery for each. - [Agentic Atlas service status](https://agentic-atlas.dev/status) (HTML) — Every refusal, in the words you received it, with what to do next. Stated copy: it reads no database, so it answers during the outage it describes. - [Agentic Atlas liveness probe](https://agentic-atlas.dev/healthz) (JSON) — `200` whenever the process is running. Reads no database, so it still answers when nothing else does. - [Agentic Atlas readiness probe](https://agentic-atlas.dev/readyz) (JSON) — `200` when this deployment can serve and `503` when it cannot, with every check, the active Release and its revision named individually. ## Agentic Atlas manifests and policy documents What this publisher is for, which addresses it serves, and on what terms. Each of these is deployment copy that reads no database, so it answers during an outage — which is when a caller most needs to be told where else to look. The three section manifests and the sitemap are the exceptions: each is an index of what the active Release publishes, so each reads it and names it. - [Agentic Atlas agent manifest](https://agentic-atlas.dev/llms.txt) (Markdown) — What this publisher is for, when to call it, and when not to. - [Agentic Atlas Nodes manifest](https://agentic-atlas.dev/nodes/llms.txt) (Markdown) — The `llms.txt` for `/nodes/`: every published Node, one link per page. - [Agentic Atlas glossary manifest](https://agentic-atlas.dev/glossary/llms.txt) (Markdown) — The `llms.txt` for `/glossary/`: every admitted term with its definition. - [Agentic Atlas decisions manifest](https://agentic-atlas.dev/decisions/llms.txt) (Markdown) — The `llms.txt` for `/decisions/`: every decision record with its ruling. - [Agentic Atlas agent signpost](https://agentic-atlas.dev/agents.md) (Markdown) — Every machine-readable surface, for a scanner that probes `/agents.md`. - [Agentic Atlas ARD manifest](https://agentic-atlas.dev/.well-known/ard.json) (JSON) — The same list as Agentic Resource Discovery entries, with a trust manifest on the MCP surfaces; the predecessor address `/.well-known/ai-catalog.json` answers the same bytes. - [Agentic Atlas API catalog](https://agentic-atlas.dev/.well-known/api-catalog) (JSON) — The APIs this host publishes, as an RFC 9727 linkset. - [Agentic Atlas pricing and usage policy](https://agentic-atlas.dev/pricing.md) (Markdown) — What the Atlas costs, what it limits, and what it licenses. - [NLWeb Schema Feeds](https://agentic-atlas.dev/schemamap.xml) (XML) — Schema.org JSON Lines feeds for Nodes, terms, and Decisions. - [Agentic Atlas sitemap](https://agentic-atlas.dev/sitemap.xml) (XML) — Every published address. - [Agentic Atlas crawler policy](https://agentic-atlas.dev/robots.txt) (Text) — Nothing is disallowed. Three tiers, each restating a content signal for what fetched bytes may be used for. ## What Agentic Atlas does not publish Stated rather than left to a failed search, because an absence you can read is cheaper than an endpoint you go hunting for. - **No write operations.** Every surface is read-only. There is no `POST` that changes anything, and therefore no idempotency key, no async job to poll, and no sandbox environment to test against. - **No webhooks and no callbacks.** Nothing here calls you back. A corpus release is published by promotion, and a client notices it by reading the revision marker on its next answer read from the Release: `X-Catalog-Revision` on a JSON API payload, a page that reads it, the sitemap, `/ask` or `/readyz`, and the coherence marker inside an MCP payload or A2A message payload. A document that reads no Release, this page included, carries none. - **No pagination.** Reads are bounded by an explicit `limit`, never windowed: there is no offset, no page number and no cursor to carry. - **No SDK and no CLI.** Consulting the Atlas needs no install of any kind — the MCP endpoint and the JSON API are both plain HTTP. The Claude Code plugin is a convenience for one client, not a dependency. - **No versioned URL prefix.** Addresses are stable and are not prefixed with `/v1/`; what changes is the Release behind them, named in `X-Catalog-Revision` on every answer read from the Release. ## If something here is wrong A surface that refuses a call it documents, or a document that describes an operation this deployment does not serve, is a bug rather than a policy. [Contact the publisher](https://agentic-atlas.dev/contact), and [why the Atlas exists](https://agentic-atlas.dev/why) is the argument for the whole thing if you would rather judge it before you build on it.