What it does
ICX is an MCP server that connects AI agents such as Claude Desktop, Cursor, VS Code and Zed to persistent memory. Agents store text, code and decisions with icx_remember, retrieve grounded facts through semantic search with icx_recall_scoped, and trace causal or dependency paths across several hops with icx_multihop_walk. It also offers sandboxed Python and WASM calculation (icx_exec, icx_wasm_exec) and shared variable state for multi-agent teams (icx_swarm_state, icx_var_set, icx_var_get).
Developers would reach for it when an agent needs to keep architecture decisions or project knowledge across sessions, or when several agents must share state without resending context as text. The publisher describes a compact context format that can replace large fact bags, and reports benchmark results on retrieval and multi-hop tasks. These figures are the publisher's own and have not been independently verified here.
The server runs as a hosted streamable HTTP or SSE endpoint, so no local model download is needed. A local stdio proxy is available for environments that only support stdio. Both options require an API key from the publisher's dashboard, and the server is added to the client's MCP configuration file. The server exposes 19 tools.
Limitations and requirements
Requires an API key from the publisher's dashboard. Uses a hosted endpoint, so the service must be reachable. The stdio proxy is an alternative for local-only environments. Claiming free tokens through the companion llm-vm-mcp server requires an email address. Benchmark results are self-reported by the publisher.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Calera Infinite Context (ICX) — Persistent Memory & Associative Recall
Sub-5ms topological 4D associative memory coprocessor for LLMs and multi-agent swarms.
v1.4.0As shown on registry.modelcontextprotocol.io when we last read it