Use RL4 with Claude Code: Cross-LLM Context in 5 Minutes

The Cross-LLM Problem

You use Cursor IDE with GPT-4 for daily coding. Then you switch to Claude Code for a complex refactor. Context? Gone. You spend 20 minutes re-explaining your architecture, decisions, and constraints.

RL4 solves this. One context, multiple LLMs, zero re-explanation.

How It Works

RL4 captures your development context in a standard format (`.rl4/` directory) and exposes it via MCP (Model Context Protocol). Any LLM that supports MCP can read the same context.

Cursor IDE ──→ RL4 MCP Server ←── Claude Code
   ↓              ↓                  ↓
Same evidence, timeline, decisions, chat history

Your context lives in one place. Both tools read from it.

Step 1: Install RL4 in Cursor IDE

If you haven't already:

Open VS Code Marketplace in Cursor
Search "RL4"
Install the extension
Open any project → RL4 creates the `.rl4/` directory automatically

After your first snapshot, you'll have evidence, timeline, and decisions captured.

Step 2: Add RL4 MCP Server to Claude Code

Claude Code supports MCP servers natively. Add RL4 to your configuration:

// In your Claude Code MCP settings (.mcp.json at project root)
{
"mcpServers": {
  "rl4": {
    "command": "node",
    "args": ["/path/to/rl4-mcp-server/dist/index.js"],
    "env": {
      "RL4_WORKSPACE": "/path/to/your/project"
    }
  }
}
}

Once configured, Claude Code has access to the same 14 MCP tools as Cursor.

Step 3: Verify the Connection

In Claude Code, try:

Search my RL4 context for recent architectural decisions

If configured correctly, Claude Code calls `search_context` and returns results from your `.rl4/` data — the same data Cursor IDE reads.

What Each Tool Does in Claude Code

All 14 MCP tools14 MCP tools/cursor/blog/rl4-mcp-tools-cursor-complete-guide work identically in Claude Code:

| Tool | Works in Claude Code? | Notes |

|------|----------------------|-------|

| `run_snapshot` | ⚠️ Cursor only | Requires VSIX extension UI |

| `search_context` | ✅ Full support | RAG search across all data |

| `rl4_ask` | ✅ Full support | Natural language Q&A |

| `search_chats` | ✅ Full support | Search Cursor chat history |

| `search_cli` | ✅ Full support | Search CLI commands |

| `get_evidence` | ✅ Full support | Read evidence pack |

| `get_timeline` | ✅ Full support | Read project timeline |

| `get_decisions` | ✅ Full support | Read structured decisions |

| `list_workspaces` | ✅ Full support | List cloud workspaces |

| `set_workspace` | ✅ Full support | Switch workspace |

| `get_content_store_index` | ✅ Full support | File index |

| `read_rl4_blob` | ✅ Full support | Read stored files |

| `rl4_guardrail` | ✅ Full support | Validate Q&A quality |

| `finalize_snapshot` | ✅ Full support | Cleanup after snapshot |

Key note: `run_snapshot` requires the Cursor VSIX extension to scan chat sources. All other tools work anywhere MCP is supported.

Real Workflow: Cursor → Claude Code

Here's a typical cross-LLM workflow:

1. Morning: Work in Cursor IDE

You spend the morning implementing a new feature in Cursor. RL4 automatically captures your chat history, file changes, and decisions.

2. Afternoon: Switch to Claude Code for a Complex Task

You need Claude's deep reasoning for an architectural decision. In Claude Code:

Ask rl4: What have I been working on today? What decisions were made?

Claude Code calls `rl4_ask` and instantly knows:

Files you modified
Conversations you had
Decisions you made
Current project state

No re-explanation needed.

3. Back to Cursor

Next session in Cursor, run a snapshot. It captures everything — including what Claude Code did while you were working there.

Cloud Sync: Context Across Machines

With RL4's Supabase sync, your context isn't limited to one machine:

Laptop (Cursor) ──→ Supabase Cloud ←── Desktop (Claude Code)

Use `list_workspaces` and `set_workspace` to access any project from any machine.

Cross-LLM Comparison: What Changes, What Doesn't

| Aspect | Same Across LLMs | Different Per LLM |

|--------|-------------------|-------------------|

| Evidence | ✅ Identical | — |

| Timeline | ✅ Identical | — |

| Decisions | ✅ Identical | — |

| Chat history | ✅ Searchable | Stored per-source |

| Skills/Rules | ✅ In `.cursor/rules/` | LLM reads differently |

| Snapshot trigger | — | Cursor VSIX only |

| MCP support | — | Native in both |

The context is identical. Only the way each LLM processes it differs — and that's the strength. Use Cursor's speed for rapid iteration, Claude Code's depth for complex reasoning, same context throughout.

Troubleshooting

"MCP server not found" in Claude Code

Verify the path to the RL4 MCP server binary
Check that `RL4_WORKSPACE` points to a directory with `.rl4/`
Restart Claude Code after config changes

"No evidence found"

Run at least one snapshot in Cursor first
Verify `.rl4/evidence.md` exists in the project
Check workspace ID matches with `list_workspaces`

Stale context in Claude Code

RL4 reads from `.rl4/` on each tool call — it's always fresh
If using cloud sync, check `last_active_at` on the workspace

Beyond Two LLMs

RL4's MCP server works with any MCP-compatible client. As the ecosystem grows:

**Codex** — OpenAI's CLI with MCP support
**Gemini CLI** — Google's CLI with MCP
**Windsurf** — MCP-compatible IDE
**Any future MCP client** — your context is ready

You're not locked into one AI. Your context is portable, structured, and always accessible.

Get Started

[Install RL4 in Cursor](/cursor/form) and create your first snapshot
Configure the MCP server in Claude Code
Try `rl4_ask` in Claude Code — see your Cursor context instantly

Learn more about the 14 MCP tools available14 MCP tools available/cursor/blog/rl4-mcp-tools-cursor-complete-guide or how to switch LLMs without losing contextswitch LLMs without losing context/cursor/blog/switch-llm-without-losing-context.

**Try RL4**Try RL4/cursor/form — one context, every LLM. Install in Cursor, use everywhere.