Tools·

from strands import Agent
from strands_emotiv.tools import BRAIN_TOOLS
agent = Agent(tools=BRAIN_TOOLS)
agent("wait until I blink twice, then tell me what my alpha did")

tool	returns
`brain_line()`	the one line, as the agent sees it
`brain_snapshot()`	the line + everything behind it
`brain_bands(channel?)`	θ α βL βH γ
`head_pose()`	yaw / pitch / roll
`contact_quality()`	14 × 0 to 4
`wait_for_brain_event(kind, timeout_s)`	blocks on blink · wink · clench · turn · `command:*`
`recent_brain_events(limit)`	the river
`brain_status()`	the doctor
`mental_approval(question)`	push the box = yes, pull = no, clench = veto
`record_start/stop/status`	cut episodes from inside a conversation
`record_publish(name?)`	push to the private Hub dataset

Every agent action calls injectMarker; the EEG record carries what the agent did and when.

Over MCP·

{ "mcpServers": { "brain": { "command": "uvx", "args": ["strands-emotiv", "mcp"] } } }

strands-emotiv mcp                 # stdio: Claude Desktop, Cursor, Kiro
strands-emotiv mcp --http 8000     # streamable HTTP at /mcp
strands-emotiv mcp --fake          # no headset

Same 13 tools, any client, own Cortex stream. Start with brain_line: prepend it to a prompt and that agent can feel the person too.