Skip to content
AI.00AI agent surface

The substrate LLM agents drive safely.

Model Context Protocol (MCP) manifest plus a workspace-scoped tool surface. Claude, ChatGPT, Gemini, Cursor and any future MCP-compatible agent can discover FlyttGo capabilities and drive platform actions under the same audit envelope a human operator would.

AI.MANDiscovery manifest

Auto-discoverable by every MCP-compatible agent.

FlyttGo publishes a Model Context Protocol manifest at the IETF-conventional /.well-known/mcp.json endpoint. Any MCP-aware client — Claude Desktop, OpenAI Agents SDK, Cursor, Continue, Cody — picks up FlyttGo automatically once the workspace operator enables it.

View live MCP manifestMCP spec ↗
curl · /.well-known/mcp.json
# Any MCP client auto-discovers FlyttGo by hitting:
curl https://flyttgotech.com/.well-known/mcp.json

# Sample fragment of the response:
{
  "name": "flyttgo-platform",
  "version": "1.0.0",
  "endpoint": {
    "transport": "http",
    "url": "/api/mcp",
    "auth": "OAuth 2.1 + workspace token"
  },
  "tools": [
    { "name": "transify.routeOrder", ... },
    { "name": "identra.verifyIdentity", ... },
    { "name": "payvera.createPaymentIntent", ... },
    { "name": "civitas.submitApplication", ... }
  ],
  "resources": [
    { "uri": "flyttgo://docs/platforms" },
    { "uri": "flyttgo://docs/deployment" },
    { "uri": "flyttgo://docs/compliance" }
  ]
}
AI.PRAgent safety pillars

Agents drive the platform under the same rails as humans.

Four guarantees that make institutional buyers comfortable handing the keys to an LLM. Each is enforced at the gateway, not in the prompt.

  • AI.PR.01

    Workspace-scoped agent tokens

    Every agent operates under a workspace token bound to a specific tenant. Tokens carry an explicit tool allowlist; calls outside the allowlist are rejected at the gateway, before the platform module sees them.

  • AI.PR.02

    Same audit envelope as humans

    Agent calls land in the same append-only JSONB before/after audit_log every human mutation does. The actor is recorded as the workspace token + agent ID, never as anonymous. Auditors can replay any agent run end-to-end.

  • AI.PR.03

    Per-tool rate + abuse detection

    Per-tool QPS caps, per-tenant daily ceilings, anomaly detection on call patterns. Suspicious sessions throttle then suspend automatically; the security desk reviews suspended sessions within one business day.

  • AI.PR.04

    Human-in-the-loop tier

    High-impact tools (payment writes, identity issuance, government submissions) carry a HITL flag. The agent calls; the platform queues a confirmation request for a designated human; only on approval does the call execute.

AI.TLAgent-callable tool registry

Six tools at preview, full surface in flight.

  • AI.T01
    platform.list
    Platform

    List the eight platforms (Transify, Workverge, Civitas, EduPro, Identra, Payvera, Ledgera, FlyttGo) with category and status.

    Scope
    public:read
    { "tool": "platform.list" }
  • AI.T02
    transify.routeOrder
    Transify

    Submit a routed order to the mobility platform. Constraints, capacity envelope and SLA target follow the workspace policy.

    Scope
    transify:write · workspace-scoped
    { "tool": "transify.routeOrder", "input": { "origin": "...", "destination": "..." } }
  • AI.T03
    identra.verifyIdentity
    Identra

    Trigger an identity verification (LoA-substantial). Returns a session URL the citizen completes; webhook delivers the result.

    Scope
    identra:write · workspace-scoped
    { "tool": "identra.verifyIdentity", "input": { "subject": "...", "loa": "substantial" } }
  • AI.T04
    payvera.createPaymentIntent
    Payvera

    Create a payment intent across SEPA, SEPA Instant, national rails or Stripe. PSD2 SCA flows applied per regulator.

    Scope
    payvera:write · workspace-scoped
    { "tool": "payvera.createPaymentIntent", "input": { "amount": 12000, "currency": "EUR" } }
  • AI.T05
    civitas.submitApplication
    Civitas

    Submit a citizen application against a Civitas service template. Routed to the responsible agency under audit envelope.

    Scope
    civitas:write · workspace-scoped
    { "tool": "civitas.submitApplication", "input": { "templateId": "...", "fields": { ... } } }
  • AI.T06
    docs.search
    Platform

    Search public documentation by free-text query. Returns top-N matching entries with source URLs.

    Scope
    public:read
    { "tool": "docs.search", "input": { "query": "sovereign deployment KSA", "top": 5 } }
AI.INIntegration patterns

Wire FlyttGo into any agent runtime.

The MCP transport works the same way across runtimes. Claude Desktop and Cursor pick it up declaratively; the OpenAI Agents SDK and Anthropic SDK wire it through one constructor argument.

  • Anthropic Claude · MCP-native
  • OpenAI Agents SDK · MCP adapter
  • Cursor + Continue + Cody · MCP client built-in
  • Custom runtime · HTTP transport · OAuth 2.1 token
typescript · @anthropic-ai/sdk + flyttgo MCP
import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic();

// Add the FlyttGo MCP server alongside any other agent tools
const message = await client.messages.create({
  model: 'claude-sonnet-4-6',
  max_tokens: 1024,
  mcp_servers: [
    {
      type: 'url',
      url: 'https://flyttgotech.com/api/mcp',
      name: 'flyttgo',
      authorization_token: process.env.FLYTTGO_WORKSPACE_TOKEN,
    },
  ],
  messages: [
    {
      role: 'user',
      content:
        'Route a same-day order from Oslo to Stockholm for ' +
        'a 2-pallet load and report back the assigned provider.',
    },
  ],
});

console.log(message.content);
// → Agent picks transify.routeOrder, applies workspace policy,
//   posts the order, and returns the routed provider summary.
AI.RMAgent surface roadmap
  • AI.RM.Q2

    Public preview · Q2 2026

    Six tools live on the MCP transport. Workspace token issuance via the trust desk; sandbox tenants seeded with synthetic data.

  • AI.RM.Q3

    Full surface · Q3 2026

    All eight platform modules expose their public-facing tool set. HITL queue available for high-impact tools. Per-tool rate caps + abuse detection in production.

  • AI.RM.Q4

    Agent governance · Q4 2026

    AIBOM (AI Bill of Materials) per workspace, signed agent identity (verifiable agent attestation), EU AI Act risk-tier classification per tool.

AI.NXWhere the agent surface plugs in

Agent access is one shape of programmatic platform access.

Agents land in the same audit envelope as the API; the same trust artefacts apply; the same procurement framework still gates production. The four pathways below take an interested team from MCP discovery to a signed deployment.