Spectra is invokable from any agent that speaks MCP.

args { action: string, target: string, task: string }

Submit one patient-portal action (one of select_practitioner, select_slot, fill_phone, submit_booking_form, fill_2fa, verify_2fa, switch_tab, open_lab_panel, read_aloud, select_medication, select_pharmacy, submit_refill, done, error). Returns the booking confirmation envelope. Proxies the Spectra frontend /api/demo/agent-step route.

args { session_id: string, description: string }

Click a named UI element in an active Spectra session. v0.2: requires the WebSocket-to-MCP bridge currently on the spectra-mcp roadmap.

args { session_id: string, field_description: string, text: string }

Type text into a named form field in an active session. v0.2 (requires WebSocket bridge).

args { session_id: string, url: string }

Navigate the active session's tab to a URL. v0.2 (requires WebSocket bridge).

args { session_id: string, direction: "up"|"down"|"top"|"bottom", amount?: number }

Scroll the active session's viewport. v0.2 (requires WebSocket bridge).

args { type: "seizure"|"aura"|"recovery"|"medication"|"note", freeform: string, duration_sec?: number, occurred_at?: ISO8601 }

Record one voice-diary entry. Returns a FHIR R4 Observation suitable for hand-off into a clinical record. Freeform is NFC-normalised server-side.

args { url: string }

Structured description of a public web page: title, headings, interactive elements with CSS selectors and importance tiers. Calls the Spectra overlay endpoint.

args { url: string }

All headings on a public page in document order with level and selector. Useful for outline navigation and auditing heading hierarchy.

args { url: string }

All interactive inputs on a public page with label, type, selector, and placeholder. Helpful for a peer agent that wants to know what a form expects before submitting it.

args { session_id: string, mode: "simplify" | "chunks" | "slow_speech" }

Switch a session into a reduced-cognitive-load mode (shorter sentences, slower speech, one step at a time). v0.2.

args { url: string }

WCAG 2.1 AA check on a public page. Returns issues with criterion references and severity. v0.2: needs a second pass with axe-core or a vision model on a screenshot.

args { region: "uk"|"ie"|"us"|"ca"|"au"|"eu"|"global", line_type?: string }

Pure directory lookup for crisis and mental-health helplines. No AI in the path; the caller decides which line is appropriate. Spectra never classifies whether a user is in crisis.

args { category?: "Navigation"|"Reading"|"Interaction"|"Modes"|"Help" }

Return the voice commands a user can say to Spectra, grouped by category. For help surfaces or for an upstream agent coaching a new user.

args { mode?: "online" | "offline" }

Language coverage for both cloud (Gemini Live, broader) and offline (Gemma 4 + Piper TTS, narrower) modes so a caller can route a user to the right provider.

args { user_group?: "motor_disability"|"deaf_mute"|"low_vision"|"cognitive_load"|"noise_constrained"|"all" }

Hand-gesture vocabulary recognised by MediaPipe Hands. Camera frames never leave the device; the MCP layer only sees the recognised intent label.

args { command: string }

Map a voice command to its gesture equivalent (or vice versa). Useful for users who have temporarily lost voice (post-surgery, noise) or permanently rely on gesture.

args { user_id: string, mode?: "gemini" | "ollama" }

Initialise a new Spectra streaming session and return the WebSocket URL the caller connects to. Either Gemini Live (cloud) or Gemma 4 nano via Ollama (offline).

args { session_id: string }

Read connection state, screen-share status, and last-activity timestamp of an active streaming session.

args { session_id: string }

Cleanly terminate a Spectra session and release its resources.