Connect to your Mind

This guide uses the daemon's HTTP tool interface. It works independently of any model provider and makes the request and result shape explicit.

You need an existing dedicated Mind endpoint and a valid capability issued for that instance. The capability must allow the tool's action and memory keyspace. Do not use a key generated by a UI demonstration or treat a timed connection animation as credential verification.

Set MINDS_INSTANCE_URL to the returned HTTPS endpoint. Supply MINDS_CAPABILITY through your application's secret configuration. If the deployment separately requires a legacy JWT, supply it in an additional Authorization: Bearer ... header; it does not replace the capability header.

1. Discover the tools

curl --fail-with-body \
  --request POST \
  --header 'Content-Type: application/json' \
  --header "x-akasha-capability: $MINDS_CAPABILITY" \
  --data '{}' \
  "$MINDS_INSTANCE_URL/v1/mcp/tools/list"

The result contains a tools array. Use each tool's advertised input schema rather than assuming every deployment enables the same list. A 401 or 403 is a credential/scope problem; it is not proof that the memory service is empty.

2. Save a small memory

curl --fail-with-body \
  --request POST \
  --header 'Content-Type: application/json' \
  --header "x-akasha-capability: $MINDS_CAPABILITY" \
  --data '{"name":"remember","arguments":{"content":"The launch review is on Tuesday.","context":{"source":"integration-test","tags":["quickstart"]}}}' \
  "$MINDS_INSTANCE_URL/v1/mcp/tools/call"

The tool output has content, is_error, and result. Check is_error before reading result. On success, the remember result reports whether storage occurred and includes a memory identifier. Save the actual returned ID; do not assume one from this example.

3. Ask for it in a later request

curl --fail-with-body \
  --request POST \
  --header 'Content-Type: application/json' \
  --header "x-akasha-capability: $MINDS_CAPABILITY" \
  --data '{"name":"recall","arguments":{"query":"When is the launch review?","limit":5,"filters":{"tags":["quickstart"]}}}' \
  "$MINDS_INSTANCE_URL/v1/mcp/tools/call"

Read result.memories, not a top-level memories field. Compare the retrieved content and source with the test record. Empty matches are a valid result; verify the same endpoint, namespace authority, and filters before changing the query.

4. Add it to your application

A typical application retrieves relevant memory before composing a model request, then stores selected outcomes after the work completes. The application decides what is worth retaining. Store source information and distinguish a user-provided fact from a model's inference.

For cleanup, inspect the deployed forget schema and target the returned test-memory ID. Its reason field is required. Avoid broad query-based deletion while validating an integration.

What this verifies

A successful save followed by a later recall verifies that these two tool requests worked against the selected instance. It does not prove backup recovery, plan billing, every SDK method, or every engine service. Those need their own workflow checks.

Source: akasha-daemon/src/main.rs routes /v1/mcp/tools/list and /v1/mcp/tools/call, akasha-daemon/src/mcp/handlers.rs, and akasha/mcp/src/tools/{mod,remember,recall,forget}.rs. Capability enforcement is in the daemon authentication middleware.