allsource-prime v0.21.3 · MCP

Memory for Claude, where your agents already work.

AllSource Prime is the AI-native memory layer for Claude Desktop, Claude Code, and any MCP client. Knowledge graph, vector recall, and compressed-index auto-injection — no CMS chrome to babysit, no embedding API keys to rotate. Your agents read and write it directly.

Install in 30 seconds

One cargo install and one config edit. Restart your client. Done.

1. Install the binary
cargo install allsource-prime
2a. Claude Desktop — ~/.claude/claude_desktop_config.json
{
  "mcpServers": {
    "prime": {
      "command": "allsource-prime",
      "args": [
        "--data-dir", "~/.prime/memory",
        "--auto-inject"
      ]
    }
  }
}
2b. Claude Code — one command
claude mcp add prime allsource-prime \
  --data-dir ~/.prime/memory \
  --auto-inject
Why this works. The --auto-inject flag exposes a prime://auto-contextMCP resource. Claude reads it at conversation start — a compressed markdown index of everything you've told it, organized by domain, in ~500–1,000 tokens. No prompt engineering required.

13 MCP tools Claude picks automatically

You don't call these. Claude does — the tool descriptions are written for agent consumption. v0.21.3 adds text-only inputs for prime_embed and prime_recall so the model never has to hand over a precomputed vector.

ToolWhen Claude calls it
prime_add_nodeCreate a node — person, concept, project, decision, insight.
prime_add_edgeConnect two nodes with a directed relation (works_on, impacts, depends_on…).
prime_embedMake a node findable by meaning. Pass text — Prime embeds it server-side via fastembed.
prime_recallHybrid recall: vector similarity + graph proximity + temporal recency. Pass text or a vector.
prime_contextCombined retrieval with tiered depth (L0 stats / L1 conversation / L2 full hybrid).
prime_indexCompressed knowledge index — token-efficient markdown of everything you know, by domain.
prime_neighborsWalk the graph around a node. Multi-hop BFS with direction + relation filters.
prime_searchList all nodes of a given type. Cheap, broad.
prime_shortest_pathFind how two entities are connected — the chain of relationships between them.
prime_similarFind the most semantically similar embeddings to a stored vector.
prime_historyFull audit trail for any entity — every creation, update, and deletion.
prime_forgetSoft-delete a node. Invisible to queries, preserved in history.
prime_statsGraph overview: total nodes, edges, types, relations.

Why not just use a CMS?

CMSes and CRMs are built for humans editing pages. Prime is built for AI agents reading and writing memory. Different shape, different wedge — pick the one your primary "editor" actually is.

FeaturePrimefolkNotionMem0
Compressed index auto-injected at conversation start
Event-sourced — time-travel and full audit trail
Cross-domain hybrid recall (graph + vector + recency)
Prime treats domains as first-class
In-process embeddings — no external API
MCP-native (Claude Desktop, Claude Code, Cursor)
Polished UI for human editors
Prime is for AI editors — that's the wedge

Comparison reflects features as of 0.21.3. folk and Notion both ship polished editing UIs — pick them if humans are still your primary writers.

What's under the hood

Prime is an AllSource event store, not a separate database. Every mutation — node, edge, vector, soft-delete — appends an immutable event to the WAL and is replayed by projections that hold the queryable state. WAL + Parquet durability with CRC32 checksums; Snappy-compressed columnar snapshots; in-memory DashMap reads at ~12µs.

Embeddings are computed in-process via fastembed (AllMiniLML6V2, 384 dims, ~25 MB, auto-downloaded into the fastembed cache on first call). No external embedding service is required. Subsequent embeddings take ~1–3 ms on a modern CPU.

Already running Claude?

Then you're halfway there. Install allsource-prime, add the MCP config block, restart your client. Claude starts writing to memory on its own — no prompt changes needed.

Back to install