Every AI. Every IDE.

Anthropic's official CLI. ai-memory runs as a stdio MCP subprocess; the memory_session_start hook fires every conversation start, returning recent-context-aware memories.

# claude mcp add
claude mcp add ai-memory --scope user --transport stdio -- ai-memory mcp --tier autonomous

VS Code-fork IDE with native MCP support. ai-memory autodiscovered. Add via Settings > Tools & MCP, restart for green-dot status indicator.

# ~/.cursor/mcp.json
{
  "mcpServers": {
    "ai-memory": {
      "command": "ai-memory",
      "args": ["mcp", "--tier", "semantic"]
    }
  }
}

First-party fork of xAI's Grok CLI maintained by AlphaOne. Native ai-memory autoload. Memory feels native; no extra config.

# Built-in. Just install grok-cli; ai-memory tools appear.
grok --memory-tier autonomous

OpenAI's developer-CLI. TOML config with underscored mcp_servers key. Use /mcp in TUI to view server status.

# ~/.codex/config.toml (or %APPDATA% on Windows)
[mcp_servers.ai_memory]
command = "ai-memory"
args = ["mcp", "--tier", "semantic"]
startup_timeout_sec = 10

Codeium's agentic IDE. Settings > Cascade > MCP Servers. ${env:VAR} interpolation supported. 100-tool limit.

# ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "ai-memory": {
      "command": "ai-memory",
      "args": ["mcp", "--tier", "smart"]
    }
  }
}

VS Code / JetBrains agent. MCP only in agent mode. Project-level .continue/mcpServers/ directory auto-detects sibling configs.

# .continue/mcpServers/ai-memory.json
{
  "name": "ai-memory",
  "command": "ai-memory",
  "args": ["mcp"]
}

Google's CLI. Use hyphens (no underscores) in server names. Tools auto-prefix as mcp_memory_*. Sanitizes *TOKEN*/*SECRET* env unless declared.

# ~/.config/gemini/settings.json
{
  "mcpServers": {
    "ai-memory": {
      "command": "ai-memory",
      "args": ["mcp", "--tier", "semantic"],
      "trust": true
    }
  }
}

AlphaOne's 24/7 multi-machine agent runner. Stacks under ai-memory for "behaviorally autonomous" deployments. Cross-host federation native.

# openclaw.toml
[memory]
backend = "ai-memory"
endpoint = "https://memory.local:9077"
mtls_cert = "/etc/openclaw/peer.pem"

Auto-generated-skill agent framework. Uses ai-memory as its long-term store; skills can reference memories by ID across sessions.

# hermes.yml
memory:
  driver: ai-memory-mcp
  tier: autonomous

Direct xAI API integration via the HTTP daemon. Authorization header passes through; per-agent_id quotas enforced server-side (v0.7.0).

# Reach the HTTP daemon directly
curl https://memory.local:9077/api/v1/recall \
  -H "x-api-key: $MEMORY_KEY" \
  -H "x-agent-id: ai:grok@host:pid-1"

Local LLM hosts. Post-#1067 (v0.7.0), any of 15 OpenAI-compatible vendors or local Ollama works; Ollama shown here for zero-network installs. No GPU required — on CPU-only or air-gapped hosts, point [llm] at a remote cloud API or internal HA endpoint instead (no local model). See the No GPU required walkthrough.

# Smart/autonomous tier — any LLM backend post-#1067; default Ollama model is gemma3:4b
ollama pull gemma3:4b
ai-memory mcp --tier smart

No native MCP, but ai-memory's HTTP daemon is reachable. Use a custom GPT with Action pointing at /api/v1. Local-only; tunnel optional.

# Custom GPT Action OpenAPI spec points to:
servers:
  - url: https://your-tunnel.example.com/api/v1

Any host implementing the Model Context Protocol v2026-04 or later auto-discovers ai-memory tools — 74 advertised entries at --profile full, 7 at --profile core. No special integration required.

# Standard MCP launcher signature
ai-memory mcp [--tier <keyword|semantic|smart|autonomous>]

Roll-your-own agents talk to the HTTP daemon. 89 routes (75 unique paths). Or implement an MCP host shim using the official spec libraries.

# Python example
import requests
r = requests.post("http://localhost:9077/api/v1/recall",
  json={"context": "...query..."})