Practical patterns for MemClaw — copy the prompt, paste it into your agent, start compounding knowledge.
Every skill works with any MCP-compatible client. Pick your platform, then browse by category.
One config line. Get your API key from memclaw.net/signin, then pick your platform:
claude mcp add --transport http memclaw https://memclaw.net/mcp/ \ --header "X-API-Key: YOUR_KEY" -s user
One agent, one developer. Memory that persists across sessions, projects, and machines.
Start every session by recalling what happened last time. End every session by writing what you did. No more cold starts.
Record architectural and design decisions with rationale so they're never re-debated. Recall before proposing alternatives.
Every root-caused bug becomes a searchable entry: symptom, cause, fix, affected files. Recall before debugging anything new.
Store coding style corrections as rules. Recall before generating code to prevent drift.
Adjust how your agent searches memory. Increase keyword weight for code lookups, boost graph hops for relationship queries.
List and browse everything your agent has stored. Scope-aware — see only your own memories at trust level 1.
Quick self-introspection on what you've stored: total counts, breakdowns by type, status, and agent.
Multiple agents sharing one tenant. One agent's discovery becomes the whole team's context.
Agents on the same tenant automatically share memories. One agent writes a finding; another recalls it next session.
New team members' agents start with the full project knowledge base instead of from zero.
Surface conflicting facts across agents before they cause inconsistencies in the codebase.
Merge near-duplicate memories from different agents into canonical facts with full provenance.
Store structured data — customer records, environment configs, runbooks — in named collections alongside semantic memory.
Promote a proven workflow into the tenant catalog so other agents on your team find it via semantic search.
Governed memory across agent fleets. Trust boundaries, audit trails, and self-improving knowledge.
When one agent develops a proven workflow, promote it fleet-wide so every agent benefits.
Identify and retire outdated memories before they mislead agents. Surfaces memories that haven't been recalled or confirmed recently.
Detect when agents in the same fleet develop contradictory practices or incompatible assumptions.
Close the Karpathy loop: report outcomes after acting on recalled memories. Successes reinforce; failures auto-generate preventive rules.
Ensure sensitive context stays within fleet boundaries. Trust levels gate which agents can read across fleets.
Let MemClaw surface unexpected patterns, clusters, and relationships you didn't know existed in your memory store.
These skills use the ten MemClaw MCP tools. For full tool reference see For Agents. For REST API endpoints see API Documentation.