Skip to content
kmem

Connect OpenClaw, Hermes or DeepSeek Harness

These three agent harnesses connect to kmem over MCP with a token. Each one keeps its settings in a different file.

The Connect screen marks all three Not yet tested with kmem. The steps on this page come from the documentation of each harness.

For each harness, create a connection first:

  1. In kmem, select Connect. The MCP tab opens.
  2. Select the harness in the list of clients.
  3. Fill in the New connection form, and select Create connection.
  4. Copy the token now. kmem shows it only one time.

Create a connection explains each field.

OpenClaw

OpenClaw keeps its settings in ~/.openclaw/openclaw.json. The file uses JSON5, which allows comments and a comma after the last item. OpenClaw reads the token from the variable KMEM_TOKEN, so the file holds no token.

  1. Under Configuration for OpenClaw, select Copy configuration.
  2. Merge the configuration into ~/.openclaw/openclaw.json. If the file already has mcp.servers, add only the kmem entry. Keep your other servers.
  3. Set KMEM_TOKEN to the token of the connection, in the environment that starts OpenClaw. For example, add export KMEM_TOKEN="YOUR_TOKEN" to your shell profile.
  4. Start OpenClaw again.
  5. Run openclaw mcp probe kmem. The command connects to kmem and counts its tools.
  6. Send the check prompt.

The configuration looks like this:

{
  mcp: {
    servers: {
      kmem: {
        url: "https://acme.kmem.app/mcp",
        transport: "streamable-http",
        headers: { Authorization: "Bearer ${KMEM_TOKEN}" },
      },
    },
  },
}

Replace acme with your org's subdomain. The Connect screen fills it in for you. Keep ${KMEM_TOKEN} as it is. OpenClaw puts the value of the variable there.

If the probe finds no tools, check two things. The environment of OpenClaw must have a value in KMEM_TOKEN, and the connection must be active.

Hermes

Hermes keeps its MCP servers in ~/.hermes/config.yaml.

  1. Under Configuration for Hermes, select Copy configuration.
  2. Merge the configuration into ~/.hermes/config.yaml. If the file already has mcp_servers, add only the kmem entry below it. Keep your other servers.
  3. In Hermes, run /reload-mcp.
  4. Send the check prompt.

The configuration looks like this:

mcp_servers:
  kmem:
    url: "https://acme.kmem.app/mcp"
    headers:
      Authorization: "Bearer YOUR_TOKEN"

Replace acme with your org's subdomain and YOUR_TOKEN with the token of your connection. The Connect screen fills in both for you.

YAML uses spaces for its structure. Keep the indentation exactly as it is. Do not use tabs.

The file holds your token. Keep it private.

DeepSeek Harness

DeepSeek Harness reads MCP servers through its MCP client plugin. You add kmem as an entry in the patch file $DSH_HOME/cordis.patch.yml. DeepSeek Harness reads the token from the variable KMEM_TOKEN.

  1. Under Configuration for DeepSeek Harness, select Copy configuration.
  2. Merge the entry into $DSH_HOME/cordis.patch.yml. If the file already holds a list, add the entry to that list.
  3. Set KMEM_TOKEN to the token of the connection, in the environment that starts DeepSeek Harness.
  4. Start DeepSeek Harness again, for example with dsh web. It reads the patch file when it starts.
  5. Open a new session.
  6. Send the check prompt.

The entry looks like this:

- insert:
    - id: mcp-kmem
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: kmem
        transport: streamable-http
        url: "https://acme.kmem.app/mcp"
        headers:
          Authorization: !!js "'Bearer ' + process.env.KMEM_TOKEN"

Replace acme with your org's subdomain. The Connect screen fills it in for you. The last line builds the header from KMEM_TOKEN when DeepSeek Harness starts. Keep it as it is.

Check each harness

Send this check prompt in a new session of the harness:

Use kmem to list the docs I can access. If there are none, say so.

Then, in kmem, select Check connection. The three checks must show Complete. Read Check a connection.

Next steps