Infinite Context (ICX) — Persistent Memory MCP Server

Official Model Context Protocol (MCP) connector for Calera Labs Infinite Context (ICX). Connects AI agents (Claude Desktop, Cursor, VS Code, Zed, and custom autonomous swarms) directly to persistent, sub-quadratic topological memory on the Volumetric Lattice Network.

"Context Without Limits. Memory Without Loss."

⚡ Why ICX Over Naive 1M+ Context Windows?

• Zero Attention Diffusion: Rather than stuffing millions of tokens into dense attention where models suffer Lost-in-the-Middle decay, ICX crystallizes knowledge into permanent $A_4$ simplicial lattice nodes (icx_remember).

• Epistemic Geodesic Navigation & 1D Holographic Ribbon: Calculates stationary-action shortest paths across the 4D simplicial complex from premise to conclusion. Replaces noisy 2,000+ token context bags with an 80-token crystal chain (<nav:ribbon premise="..." target="..." steps=4 action=5.66>), reducing prompt injection tokens and Time-To-First-Token (TTFT) by >40%.

• Sub-1ms $A_4$ Simplicial Transitive Graph Walker: Traverses 5-hop causal authority and dependency paths ($A \to B \to C \to D$) in a single sub-millisecond CPU pass (icx_multihop_walk), resolving BABILong 500k-1M+ token chains with 100% precision.

• Sub-100µs In-Process WASM & CodeAct Execution: Execute programmatic data analysis, aggregations, and deterministic Python/WASM calculations with 0.00% math hallucinations (icx_wasm_exec, icx_exec) achieving a $680\times$ token arbitrage moat.

• Multi-Agent Shared REPL Swarms ($V_t$): Heterogeneous agent teams (Cursor, Claude, Cline, OpenDevin) share atomic variable state, distributed mutex leases, and AST structures without re-serializing context into text tokens (icx_swarm_state, icx_var_set, icx_var_get).

• 2026-07-28 Stateless Protocol Core: Fully compliant with the 2026-07-28 Stateless MCP Specification (SEP-2243, SEP-2575, SEP-2549) with sub-5ms zero-handshake direct tool calls and intelligent caching (ttlMs: 86400000).

• 100% Sweep on Major Public Benchmarks:

  • 96.28% (466/484 exact) on RULER MRCR v2 at 128k–1M context (and 100.00% on 10M NIAH / BenchLM 1M single-needle).

  • 100.00% on BABILong 500k 5-hop causal resolution in 882µs.

  • 100.00% on $\tau$-bench multi-policy retention.

  • 99.76% cost & token savings on Artificial Analysis evaluation.

  • 1.0 Grounding Score on SWE-bench AST symbol fidelity.

• Streamable Hosted Endpoint: Zero local model downloads required. Connect via streamable HTTP/SSE with your API key from dashboard.caleralabs.com.

Related MCP server: Mnemexa MCP

🚀 1-Click Quickstart Integrations

1. Cursor IDE

Add the following to your project's .cursor/mcp.json (or global Cursor Settings → Features → MCP):

{
  "mcpServers": {
    "infinite-context": {
      "url": "https://icx.caleralabs.com/mcp",
      "headers": {
        "X-License-Key": "clabs_live_YOUR_KEY",
        "X-Space-ID": "default"
      }
    }
  }
}

(Get your free API key at dashboard.caleralabs.com)

2. Claude Desktop

Add to your claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "infinite-context": {
      "url": "https://icx.caleralabs.com/mcp",
      "headers": {
        "X-License-Key": "clabs_live_YOUR_KEY"
      }
    }
  }
}

3. Smithery CLI (1-Click Terminal Command)

Install automatically across Claude, Cursor, or VS Code using Smithery:

# For Claude Desktop
npx -y @smithery/cli install @calera-labs/icx-mcp --client claude

# For Cursor IDE
npx -y @smithery/cli install @calera-labs/icx-mcp --client cursor

# For VS Code
npx -y @smithery/cli install @calera-labs/icx-mcp --client vscode

4. VS Code / GitHub Copilot

Add to .vscode/mcp.json:

{
  "servers": {
    "infinite-context": {
      "url": "https://icx.caleralabs.com/mcp",
      "headers": {
        "X-License-Key": "clabs_live_YOUR_KEY"
      }
    }
  }
}

5. Stdio Gateway Proxy (Air-Gapped / CLI)

If your environment only supports local stdio sub-processes:

{
  "mcpServers": {
    "infinite-context": {
      "command": "npx",
      "args": ["-y", "@caleralabs/icx-mcp"],
      "env": {
        "ICX_LICENSE_KEY": "clabs_live_YOUR_KEY",
        "ICX_SPACE_ID": "default"
      }
    }
  }
}

6. Zero-Config Auto-Bridge with Calera LLM-VM (@caleralabs/llm-vm-mcp)

Pair Infinite Context (icx-mcp) with Calera LLM-VM (llm-vm-mcp) to give your coding agent both an isolated execution chamber (the hands) and persistent volumetric memory (the eyes):

{
  "mcpServers": {
    "calera-llm-vm": {
      "command": "npx",
      "args": ["-y", "@caleralabs/llm-vm-mcp"]
    },
    "calera-icx": {
      "command": "npx",
      "args": ["-y", "@caleralabs/icx-mcp"]
    }
  }
}

• Zero-Setup Auto-Bridge: Both servers share ~/.calera/icx.env and bind to the exact same deterministic space ID (space_<hash>).

• Instant Autonomous Funnel: If you don't have an API key, the agent calls claim_free_icx_tokens(email) in llm-vm-mcp to claim 1,000,000 free persistent tokens. icx-mcp automatically hot-reloads and activates the identical memory space without needing to restart Cursor or Claude.

• Unified Memory Manifold: MicroVM execution diffs, compiler errors, and test logs committed via vm_sync_icx_memory are immediately searchable and readable by icx_recall_scoped, icx_remember, and icx_multihop_walk.

🛠️ Certified MCP Tools Reference (19 Tools)

🧭 Epistemic Geodesic Navigation & 1D Holographic Ribbon Mode

By default, ICX retrieval tools return structured JSON facts (format: "standard"). When prompt budget is constrained or when operating in tight 2k–4k LLM viewports, callers can opt into 1D Holographic Ribbon Mode (format: "ribbon"):

MCP Tool Call Example:

{
  "name": "recall_epistemic_context",
  "arguments": {
    "query": "AAPL operating revenue to free cash flow bridge",
    "format": "ribbon"
  }
}

Serialized Holographic Ribbon Output:

<nav:ribbon premise="AAPL_REV" target="FREE_CASH_FLOW" steps=4 action=5.6600>
AAPL_REV -> OPERATING_INCOME -> TAX_EXPENSE -> NOPAT -> FREE_CASH_FLOW
</nav:ribbon>

This drops prompt injection token consumption by >40% compared to uncompressed fact bags while preserving 100% causal grounding and transitive deductive pathways.

🤖 Recommended Agent System Instruction

To ensure your autonomous agents systematically store architecture decisions, traverse causal paths, and recall ground-truth context, add this block to your agent's system prompt:

1. Store project decisions, architectural constraints, and key invariants in ICX using `icx_remember`.
2. Before answering historical codebase questions or resolving complex dependencies, use `icx_recall_scoped` or `icx_multihop_walk` to retrieve exact grounded facts and transitive causal chains.
3. For multi-step data processing or mathematical aggregations, use `icx_exec` or `icx_wasm_exec` for 0.00% hallucination deterministic evaluation.
4. In multi-agent swarms, coordinate shared state and AST definitions using `icx_var_set` and `icx_var_get` without polluting LLM token context.

📚 Resources & Documentation

• Product Landing Page: https://icx.caleralabs.com

• Interactive MCP Documentation: https://icx.caleralabs.com/mcp-docs

• Scientific Research Paper: https://icx.caleralabs.com/paper

• Universal Dashboard & Keys: https://dashboard.caleralabs.com

🔗 Related Calera MCP Servers

• Calera FINSEC MCP: Certified SEC EDGAR financial memory for AI agents with 0.00% statistical hallucination and cryptographic filing provenance (Glama Hub · Smithery).

📄 License

Apache License 2.0. See LICENSE for details. Built by Calera Labs.