MCP Server — GEIANT Perception Service

The GEIANT perception service is a live Model Context Protocol server that exposes geospatial AI tools over SSE. Every tool call is audit-wrapped with delegation checks and breadcrumb drops.

Connection

SSE endpoint: https://packagesmcp-perception-production.up.railway.app/sse
Messages:     POST /message?sessionId=<id>
Health:       GET /health

Connect with MCP SDK

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { SSEClientTransport } from '@modelcontextprotocol/sdk/client/sse.js';

const BASE = 'https://packagesmcp-perception-production.up.railway.app';
const transport = new SSEClientTransport(new URL(`${BASE}/sse`));
const client = new Client({ name: 'my-app', version: '1.0.0' });

await client.connect(transport);
const tools = await client.listTools();
console.log(tools.tools.map(t => t.name));
// ['perception_fetch_tile', 'perception_classify',
//  'perception_embed', 'perception_weather']

Connect with Claude Desktop

Add to your Claude Desktop MCP config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "geiant-perception": {
      "url": "https://packagesmcp-perception-production.up.railway.app/sse"
    }
  }
}

Available Tools

perception_weather

Fetch historical or forecast weather for any H3 cell via Open-Meteo ERA5.

const result = await client.callTool({
  name: 'perception_weather',
  arguments: {
    h3_cell: '851e8053fffffff',          // Rome
    timestamp: '2026-03-20T12:00:00Z',   // Optional, defaults to now
    write_to_spatial_memory: false,       // Write result to Supabase
  },
});

const weather = JSON.parse(result.content[0].text);
// { temperature_c: 18.1, wind_speed_ms: 1.57, status: 'ok', ... }

perception_fetch_tile

Fetch Sentinel-2 L2A satellite tiles for any H3 cell. Returns STAC metadata and band URLs.

const result = await client.callTool({
  name: 'perception_fetch_tile',
  arguments: {
    h3_cell: '851e8053fffffff',
    max_cloud_cover: 20,
    bands: ['B02', 'B03', 'B04', 'B8A', 'B11', 'B12'],
  },
});

perception_classify

Run Prithvi-EO-2.0 foundation model inference on a fetched tile. Requires a running RunPod GPU endpoint.

const result = await client.callTool({
  name: 'perception_classify',
  arguments: {
    h3_cell: '851e8053fffffff',
    model: 'prithvi-eo-2.0-300m',
    task: 'flood_detection',
  },
});

perception_embed (Planned)

Generate Clay v1.5 geospatial embeddings for a tile. Phase 4.2 — coming soon.

HTTP Endpoints

Beyond MCP tools, the service exposes REST endpoints:

Endpoint	Method	Description
/health	GET	Service status, chain tip, audit state
/test/weather	GET	Quick test — drops a breadcrumb
/epoch/roll	POST	Roll all unrolled breadcrumbs into an epoch
/compliance	GET	Generate EU AI Act compliance report
/compliance/:agent_pk	GET	Compliance report for a specific agent

Epoch Rollup

curl -X POST https://packagesmcp-perception-production.up.railway.app/epoch/roll

{
  "success": true,
  "epoch_index": 1,
  "block_count": 3,
  "start_block": 5,
  "end_block": 7,
  "merkle_root": "6fa3e35a8f8fd9d4...",
  "tier_at_close": "provisioned",
  "epoch_hash": "5c6b317cadc39538..."
}

Compliance Report

curl https://packagesmcp-perception-production.up.railway.app/compliance

Returns a full EU AI Act compliance report with chain verification, epoch summaries, delegation certificate, trust score, and violation history.

Audit Trail

Every MCP tool call automatically:

1. Pre-flight: Checks delegation certificate (jurisdiction, facet, temporal validity, tool whitelist)
2. Executes: Runs the actual tool
3. Post-flight: Builds context digest, chains to previous block, Ed25519-signs, writes to Supabase

If pre-flight fails, the tool call is blocked — it never executes. A compliance violation is logged instead.

The chain tip is visible in the /health endpoint:

curl -s .../health | jq '{chain_tip, audit_active, agent_pk}'
# { "chain_tip": 8, "audit_active": true, "agent_pk": "c14094ea7efb6122" }