OpenClaw & Hermes

OpenClaw and Hermes both speak MCP, so mnesio plugs in the same way it does for any MCP client: point the runtime at the mnesio-mcp binary. Ready-to-copy configs live in the repo under examples/integrations/.

Build the binary

cargo build --release -p mnesio-mcp
# -> target/release/mnesio-mcp

Configure the runtime

{
  "mcpServers": {
    "mnesio": {
      "command": "/abs/path/to/mnesio/target/release/mnesio-mcp",
      "args": [],
      "env": { "MNESIO_DATA_DIR": "/abs/path/to/mnesio-data" }
    }
  }
}

{
  "mcpServers": {
    "mnesio": {
      "command": "/abs/path/to/mnesio/target/release/mnesio-mcp",
      "args": [],
      "env": { "MNESIO_DATA_DIR": "/abs/path/to/mnesio-data" }
    }
  }
}

// examples/integrations/openclaw.skill.json — teach the agent when to use memory
{
  "name": "long-term-memory",
  "description": "Persist and recall facts across sessions via mnesio.",
  "tools": ["mnesio_search", "mnesio_write_memory", "mnesio_record_outcome"],
  "guidance": "Search memory before answering anything that may depend on prior context; write salient new facts after each turn."
}

Verify

After registering, confirm the runtime can see the three tools (mnesio_write_memory, mnesio_search, mnesio_record_outcome). You can sanity check the server independently:

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' \
  | ./target/release/mnesio-mcp

See the integration guide for the general agent-loop pattern and the measured 0% → 83% result.