One memory for CrewAI, and every tool you will use
Researchers, analysts, and writers read and write the same Hypermemory, and you keep that graph when the crew hands off to Claude or another SDK.
Why This Works
Hypermemory is accessed via MCP and the CLI, the same memory graph either way. CrewAI agents call the MCP endpoint or shell out to the CLI, hm_store and hm_recall on the same nodes your other tools use. Specialist agents compound knowledge instead of duplicating it in isolated prompts.
Quick Setup
MCP URL: https://api.hypermemory.io/mcp
- Create a Hypermemory account and open app.hypermemory.io.
- Get an API key and graph ID from API Keys in the app.
- Point your CrewAI agents at the Hypermemory MCP URL with that API key.
- Add the Hypermemory skill (or hm_ tool wrappers) so every agent on the crew uses the same graph.
Available Tools
| Tool | Description |
|---|---|
| hm_find_related | Traverse connected knowledge from a starting node |
| hm_forget | Delete a node with optional cascade |
| hm_get_overview | Graph stats, top nodes, and type distribution |
| hm_ingest | Decompose dense text into graph entities in one pass |
| hm_list_files | Query uploaded files by type or name |
| hm_list_orphans | List disconnected nodes for cleanup |
| hm_recall | Hybrid search across your graph (BM25, vector, edges) |
| hm_store | Create a memory node; background enrichment runs automatically |
| hm_timeline | Temporal recall by period or date range |
| hm_timeline_write | Write an explicit diary-style timeline entry |
| hm_update | Update a node; re-embeds when the description changes |
| hm_upload_file | Upload a file into memory (summary + graph node) |
Use Case
Research stores market notes with hm_ingest; strategy writes positioning with hm_store; copy pulls hm_recall before drafting. Next week you debug the launch in Claude, those nodes are already there. No export step between CrewAI and your chat tools.