# aeon-monitor-mcp -- Full Reference # MCP server wrapping the running aeon-monitor control + data HTTP API (gnosis.monitor.v1): scene catalog, status, frame fetch, capture start/stop, tune, and source selection. ## Server - Name: aeon-monitor-mcp - Version: 1.0.0 - MCP Protocol: 2024-11-05 - Endpoint: https://aeon-monitor.forkjoin.ai/mcp - Discovery/docs quota: none - Paper: "Being Irreversible" by Taylor William Buley ## Tools ### monitor_list_scenes List the aeon-monitor scene catalog (GET /monitor/scenes). Returns [{ id, title, kind, live }] for every renderable scene (skymesh, spectrum-waterfall, noaa-apt, image, mesh, knot-viz, selftalk, residual, gate-telemetry). Input schema: ```json { "type": "object", "properties": {} } ``` ### monitor_status Get the live aeon-monitor status (GET /monitor/status, gnosis.monitor.status.v1): activeScene, source, the capture state (running, kind, centerHz, iqRateHz, gain, sourceId, lastError), and the scene catalog. Input schema: ```json { "type": "object", "properties": {} } ``` ### monitor_get_frame Fetch the current frame envelope for a scene (GET /monitor/frame.json?scene=, gnosis.monitor.v1). For raster scenes the large base64 raster is NOT returned; the result summarizes scene, kind, width, height, channels, center frequency, attributes, and metadata. Vector/embed/skymesh payloads are returned in full. Input schema: ```json { "type": "object", "properties": { "scene": { "type": "string", "description": "Scene id from the catalog (e.g. \"spectrum-waterfall\", \"noaa-apt\", \"skymesh\"). Omit to use the monitor active scene." } }, "required": [ "scene" ] } ``` ### monitor_select_scene Set the active scene the no-?scene frame/stream + UDP publisher default to (POST /monitor/control op=select_scene). Returns the resulting status. Input schema: ```json { "type": "object", "properties": { "scene": { "type": "string", "description": "Scene id from the catalog to make active." } }, "required": [ "scene" ] } ``` ### monitor_start_capture Start the managed capture producer (POST /monitor/control op=start_capture). Fail-soft: optional kind, centerHz, iqRateHz, and gain override the current capture config. Returns the resulting status. Input schema: ```json { "type": "object", "properties": { "kind": { "type": "string", "description": "Capture kind (e.g. \"spectrum\"). Defaults to the current kind." }, "centerHz": { "type": "number", "description": "Center frequency in Hz." }, "iqRateHz": { "type": "number", "description": "IQ sample rate in Hz." }, "gain": { "type": "number", "description": "Tuner gain (dB)." } } } ``` ### monitor_stop_capture Stop the managed capture producer (POST /monitor/control op=stop_capture). Returns the resulting status. Input schema: ```json { "type": "object", "properties": {} } ``` ### monitor_tune Retune the live capture to a new center frequency (POST /monitor/control op=tune). Restarts the producer at centerHz if it is running. Returns the resulting status. Input schema: ```json { "type": "object", "properties": { "centerHz": { "type": "number", "description": "New center frequency in Hz." } }, "required": [ "centerHz" ] } ``` ### monitor_set_source Select the capture source device (POST /monitor/control op=set_source). Returns the resulting status. Input schema: ```json { "type": "object", "properties": { "sourceId": { "type": "string", "description": "Source/device id to capture from." } }, "required": [ "sourceId" ] } ``` ### monitor_query Run REAL inference on a node in the Skymesh mesh (POST /monitor/query?mesh=&node=). The relay forwards the request to a bridge in the room, which runs the VERIFIED per-token decode-next path (NOT the broken batched /generate) against its local fat-station and returns the decoded reply. Provide a `prompt` to tokenize, OR raw `tokens` (the node's in-tree tokenizer is a tiny smoke table, so arbitrary prose may 400 — pass token ids as the escape hatch, e.g. [785,6722,315,9625,374]). Returns the reply text, the output token ids, the model, and the latency. When the serving bridge is cache-aware it also returns `cached` (true = served from the GLOBAL ANSWER CACHE with no inference, false = fresh inference) and `cacheTier` for a hit. Generation is slow (a few seconds) unless served from cache. On a node-side issue (no fat-station / tokenizer 400) the result includes a `detail` hint. Input schema: ```json { "type": "object", "properties": { "prompt": { "type": "string", "description": "Prompt text to tokenize on the node and run inference on. Ignored when `tokens` is given." }, "tokens": { "type": "array", "items": { "type": "number" }, "description": "Raw prompt token ids (bypass the node smoke tokenizer). Use this for arbitrary prose the in-tree tokenizer cannot encode." }, "maxTokens": { "type": "number", "description": "Max new tokens to decode (default 24)." }, "mesh": { "type": "string", "description": "Mesh id (the relay room). Omit for the global partyline (skymesh-global)." }, "node": { "type": "string", "description": "Optional node id to target a specific bridge in the room." } } } ``` ## Discovery Endpoints - /llms.txt -- This file - /llms-full.txt -- Full reference (you are here) - /skills -- Skill definitions - /agents.json -- Agent manifest - /.well-known/mcp.json -- MCP discovery - /.well-known/agent-card.json -- Agent card - /.well-known/openapi.json -- OpenAPI spec - /.well-known/security.txt -- Security contact ## Integration Connect via any MCP-compatible client (Claude, ChatGPT, Cursor, etc.) POST https://aeon-monitor.forkjoin.ai/mcp with JSON-RPC 2.0 body. ## Parent Part of the ForkJoin MCP server family at https://forkjoin.ai Superserver: https://mcp.forkjoin.ai