Cline (VS Code extension) — MCP server + custom instructions

Category 2. Cline is MCP-capable; configure via Cline’s Settings panel or ~/.cline/mcp_settings.json (varies by version).

Quick install

Cline’s MCP config path varies between releases (the file has lived at ~/.cline/mcp_settings.json and under the VS Code extension data dir), so the installer requires --config <path>:

# TODO(#487): once Cline pins a canonical path, --config will be optional.
ai-memory install cline --config ~/.cline/mcp_settings.json
ai-memory install cline --config ~/.cline/mcp_settings.json --apply
ai-memory install cline --config ~/.cline/mcp_settings.json --uninstall --apply

Find your active config by opening Cline → Settings → MCP and noting the file path it reads from. This handles Part 1 below; Part 2 (custom instructions) is still manual.

Config file location (current Cline VS Code extension)

Recent Cline releases store MCP settings under the VS Code extension’s globalStorage dir for publisher saoudrizwan.claude-dev (Cline was formerly “Claude Dev”), in cline_mcp_settings.json:

Swap Code for Code - Insiders (VS Code Insiders) or VSCodium (VSCodium) if you run those. The fastest way to find it regardless of version: Cline → Settings → MCP Servers → Configure opens this file directly. Pass it to the installer via --config <that path>.

On Windows the command must be the full .exe path and use forward slashes in JSON, e.g. "command": "C:/Users/<you>/.local/bin/ai-memory.exe" with "args": ["--db", "C:/Users/<you>/.local/share/ai-memory/memories.db", "mcp", "--tier", "smart"].

Part 1 — MCP server

{
  "mcpServers": {
    "ai-memory": {
      "command": "ai-memory",
      "args": ["mcp"],
      "env": { "AI_MEMORY_DB": "${HOME}/.claude/ai-memory.db" },
      "disabled": false,
      "autoApprove": ["memory_session_start", "memory_recall", "memory_capabilities"]
    }
  }
}

autoApprove lets the model call memory tools without prompting for permission on every call — required for a smooth boot path. The set above is the conservative read-only minimum. For seamless auto load + store across chat sessions (the common request), approve the read-mostly working set including the memory_store write:

{
  "autoApprove": [
    "memory_store", "memory_recall", "memory_search", "memory_list",
    "memory_get", "memory_load_family", "memory_smart_load", "memory_capabilities"
  ]
}

Note this includes memory_store (a write) so the model can persist context without a prompt on every turn; drop it back to the read-only set if you prefer to approve each write manually. A timeout (seconds, e.g. "timeout": 300) is also honored by the Cline extension for slower first-call model/embedder warm-up.

Using --tier smart or --tier autonomous with a non-default LLM backend? Post-#1146 (v0.7.0) the recommended path is a [llm] section in ~/.config/ai-memory/config.toml — single source of truth across MCP / HTTP daemon / CLI / boot banner / doctor probe. Example: backend = "xai", model = "grok-4.3", api_key_env = "XAI_API_KEY" (the env-var name, not the literal key — inline keys are rejected at parse time). Export the named env var in your shell rc; the MCP config can stay minimal. Override path: extend the env block above with AI_MEMORY_LLM_BACKEND, AI_MEMORY_LLM_API_KEY, and AI_MEMORY_LLM_MODEL — shell exports don’t reach MCP-spawned subprocesses (#1144). Full schema + per-vendor recipes: ../CONFIG_SCHEMA.md + llm-backends.md.

Part 2 — Custom Instructions (best-effort)

Settings → Cline → Custom Instructions:

At the start of every conversation, before responding to the user’s first message, call memory_session_start then memory_recall against the current project’s namespace. Reference recalled titles in your first reply.

Limitation

Same as Cursor (category 2). A native SessionStart hook would close the gap; cross-file at Cline upstream tracked in #487.

Related

• README.md, Issue #487