Use the catalog from your agent
What the catalog is for, then how to wire it in. Pick a use case to see how it works; below, two ways to connect — live tools over the hosted MCP server, or a drop-in system prompt for clients that can't.
What people bring to the catalog once it's connected. Expand any one for how it works and a prompt to try.
Build a new agentScaffold a system on named patterns, not ad-hoc structure.
Hand the catalog to your coding tool and let it design against the field's vocabulary. It pulls a fitting recipe, expands the patterns it composes, and gives you a structure you can defend by name — each piece carrying its intent, forces, and known liabilities — instead of one improvised from scratch.
Try
Design an agent that triages support tickets. Recommend a recipe, then expand the patterns it composes and explain why each fits.
Verify and de-risk a designSurface gaps, blind spots, and the anti-patterns you are about to ship.
Describe an existing architecture and ask which patterns it already implements and which it is missing. The same pass runs the design past the known anti-patterns and each pattern's liabilities, so failure modes turn into a pre-ship checklist rather than a production incident.
Try
Here is my agent design. Which catalog patterns does it implement, which is it missing, and which anti-patterns is it at risk of?
Diagnose by symptomName what is going wrong, get the patterns and anti-patterns that fix it.
When a running agent misbehaves — loops forever, hallucinates tool calls, blows its context window — describe the symptom and get back the patterns that address it and the anti-patterns that explain it. This is troubleshooting a live system, not a design review.
Try
My agent keeps looping on the same step and never finishes. Which patterns prevent this, and which anti-pattern am I hitting?
Decide between approachesMake a trade-off explicit by walking the typed edges between patterns.
When two patterns compete — ReAct versus plan-and-execute, single- versus multi-agent — walk the alternative-to and conflicts-with edges to see what each buys and costs. You get a decision you can write down and defend, not a coin flip.
Try
Should I use ReAct or plan-and-execute for a long-horizon coding agent? Compare them via their related edges and trade-offs.
Lift a worked exampleStart from a real implementation instead of a blank file.
Most patterns carry concrete examples — wiring, prompts, control flow as people actually run them. Ask for the examples behind a pattern and copy from a working shape rather than reconstructing it from the description.
Try
Show me worked examples of the reflection pattern, then adapt one to a code-review agent.
Speak one languageA shared, citable vocabulary for design docs, ADRs, and reviews.
Every pattern has a stable id. Cite it in a design doc, an ADR, or a pull-request review and everyone reads the same thing — the intent, the forces, the liabilities — instead of arguing past each other over homegrown names. The catalog becomes the team's common reference.
Try
Review this design doc and annotate each component with the catalog pattern id it corresponds to.
Learn the fieldCome up to speed from symptom to pattern to recipe to example.
Use the catalog as a course, not just a lookup. Follow the edges from a single pattern out to the recipes that compose it and the examples that ground it, and lean on the methodologies, glossary, books, and trainings behind them to onboard yourself or a team.
Try
I am new to agent design. Walk me from chain-of-thought up to its parent recipes, then point me at examples and a glossary term to read next.
Your agent calls live tools — find_pattern, recommend_recipe, pattern_for_symptom, glossary_term, and 19 more — against 421 patterns, 161 compositions, 54 methodologies, and 49 glossary terms. Configure once per client; the catalog stays fresh on every call.
Endpoint
https://mcp.agentpatternscatalog.org/mcp
Pick your client
One CLI call. After it returns, /mcp in any Claude Code session shows the connection.
Command
claude mcp add --transport http agent-patterns https://mcp.agentpatternscatalog.org/mcp
Verify
Ask your agent: “Find me memory patterns.” You should see a tool call and a list back — like this:
You: Find me memory patterns.
Agent: → find_pattern(query="memory", limit=5)
← - agentic-memory [memory]
- five-tier-memory-cascade [memory]
- knowledge-graph-memory [memory]
- memory-poisoning [anti-patterns]
- memory-extraction-attack [anti-patterns]