Agentic Atlas
VerifiedFeaturedRemoteAbout Agentic Atlas
Agentic Atlas is a collection of agent-system design patterns for people and agents to consult over MCP. Its Claude Code companion plugin offers skills to review designs, plan systems, and apply named patterns.
Key features
- 01Consult agent design patterns
- 02Review existing agent designs
- 03Plan new agent systems
- 04Create cited pattern-based edit plans
- 05Claude Code plugin and individual skills
Use cases
- 01Review an existing agent artifact
- 02Plan an agent system architecture
- 03Develop an edit plan from a named pattern
MCP support
Agentic Atlas provides a hosted, read-only MCP server for clients to consult patterns and related content. The README also describes connecting the Claude Code plugin to this endpoint.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://agentic-atlas.dev/mcp/claude mcp add --transport http agentic-atlas-bf5d87 https://agentic-atlas.dev/mcp/{"mcpServers":{"agentic-atlas-bf5d87":{"type":"http","url":"https://agentic-atlas.dev/mcp/"}}}Open in Cursor{"servers":{"agentic-atlas-bf5d87":{"type":"http","url":"https://agentic-atlas.dev/mcp/"}}}Open in VS Code{"mcpServers":{"agentic-atlas-bf5d87":{"type":"http","url":"https://agentic-atlas.dev/mcp/"}}}{"mcpServers":{"agentic-atlas-bf5d87":{"serverUrl":"https://agentic-atlas.dev/mcp/"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
5| Parameter | Type | Required | Description |
|---|---|---|---|
| query | string | Yes | Ordinary design vocabulary. Stopwords are dropped and each remaining word matches whole tokens (a word of four or more characters also matches longer tokens it begins) in canonical ids, titles, Hooks, sealed Card claims, and published Section headings and bodies, each word weighed by how few subjects carry it in an identity field. A Section match orders a candidate but makes one only when no subject matches in an identity field, so total counts the candidates, not every subject a word touches. With kind='decision', the searchable identity is the raw canonical id, title, and complete sealed Decision body. With kind='term', the canonical glossary key and sealed definition are searched. Empty lists every subject of the selected kind. |
| kind | any | No | Restrict to 'pattern', 'concept', 'example', or 'decision', or 'term'; any kind excludes Roadmap entries. Decision and term candidates carry directly readable addresses. |
| status | any | No | One status-ladder rung; published subjects are 'fleshed' or 'stable'. |
| bound | integer | No | Candidates per page, 1-8. |
| cursor | any | No | scope.next_cursor from a prior atlas_orient page; repeat the same other arguments. |
| expected_revision | any | No | A prior payload's 'coherence'; 'revision_changed' refuses a moved Release. |
Details
From the project READMEAgentic Atlas — companion plugin for Claude Code
Agentic Atlas is a collection of field-tested patterns for designing agent systems, readable by people and consultable by agents over MCP. This repository distributes its public Claude Code companion plugin.
Install individual skills
npx skills add aj604/agentic-atlas-plugin --skill reviewing-agent-designs
npx skills add aj604/agentic-atlas-plugin --skill designing-agent-systems
npx skills add aj604/agentic-atlas-plugin --skill applying-a-pattern
Each standard skills CLI command explicitly selects one skill —
reviewing-agent-designs audits an existing agent artifact against the
patterns, designing-agent-systems designs a new agent system, and
applying-a-pattern turns one named pattern into a cited edit plan. A
command installs that skill alone, not the repository's Claude Code plugin
manifest, MCP configuration, pattern-librarian agent, design-auditor
agent, or shared return-shapes contract; the skill then runs its
consultation method inline. Connect the hosted MCP endpoint for your
client first: https://agentic-atlas.dev/connect.
Browse the repository and its three capability areas on skills.sh:
Install the Claude Code companion plugin
/plugin marketplace add aj604/agentic-atlas-plugin
Then install agentic-atlas. The plugin connects to the hosted, read-only MCP
endpoint at https://agentic-atlas.dev/mcp/. It installs no SDK, CLI, local
server, or content bundle.
The three skills can read an agent artifact you select in your workspace. They send only pattern lookup terms, Atlas identifiers, and consultation parameters to the hosted endpoint, not your artifact text or local file paths. The plugin has no hooks, background process, or local data store. The endpoint's data handling is described in the Atlas privacy policy.
Connection instructions are at https://agentic-atlas.dev/connect, and the copyable consultation contract is at https://agentic-atlas.dev/consult.md. Report plugin or transport problems in this repository's issue tracker.
Developer resources
The Atlas is a hosted publication, and its developer surfaces are served from the domain rather than from this repository. All of them are read-only and unauthenticated, and all are indexed at https://agentic-atlas.dev/developers:
- Agentic Atlas MCP server
- Agentic Atlas HTTP API
- Agentic Atlas authentication
- Agentic Atlas availability
- Agentic Atlas manifests and policy documents
Working on the plugin with an AI coding agent? Start with AGENTS.md.
License
The plugin is distributed under the Agentic Atlas License:
source-available, free to read, install, and consult, with republication
withheld. Both manifests name it as LicenseRef-Agentic-Atlas-1.0.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.