Integrations
Technical Overview

Technical Overview

Audience: Engineers integrating with Adros through MCP, REST, webhooks, or the dashboard.

Companion docs:

1. What Adros is

Adros is a hybrid marketing OS: the same FastAPI + Postgres + MCP backend powers both operator-grade dashboard workflows and AI agent integration.

  • Marketing brain: pattern database, memory layer, creative pipeline, daily monitoring, and managed specialist workflows
  • Research stack direction: Perplexity + DataForSEO + Firecrawl + platform-native Meta/Google data
  • Operator surfaces: app.adros.ai (opens in a new tab) includes Command Center, Strategy Workspace, Files, Campaigns, and Adros CEO
  • Tasks and events are the system of record whether work starts in the UI, CEO chat, or MCP

Runtime actors (April 2026)

ActorIDRole
Adros CEOadros_ceoConversational coordinator; intake; routes work into tracked tasks
Adros Researchadros_researchManaged Agent - keyword research, market research, competitor research, and audience research
Adros CMOadros_cmoManaged Agent - audits, strategy synthesis, structure/strategy, and optimization analysis
Adros CDadros_cdCopy and creative direction
Adros Deployeradros_deployerLaunch and account change execution

2. Stack

LayerTech
BackendFastAPI on Railway (api.adros.ai)
DatabasePostgres via Supabase
File storageCloudflare R2
FrontendReact + Vite on Vercel (app.adros.ai)
MCP serverStreamable HTTP at /mcp
AuthJWT (lyr_...)
Vector searchQdrant
NotificationsResend, WhatsApp demo/BYOK paths

3. Data model (key entities)

Users & multi-tenancy

  • users
  • clients
  • client_lifecycle_state
  • client_integrations
  • Data is scoped by user_id and, where applicable, client_id

Tasks & orchestration

  • tasks - unit of work with status, specialist/agent, cost, and project linkage
  • task_events - append-only timeline
  • task_comments - discussion, with optional follow-up task creation
  • task_blockers - human-required blockers
  • projects - grouping layer
  • WebSocket at /ws/events/{task_id}?token=lyr_...

Strategy & artifacts

  • Workspace stages from research through deploy
  • Stage masters
  • Snapshots
  • Deploy requests

Patterns & memory

  • patterns
  • memory_store for business context, personas, and weekly logs

Monitoring

  • monitoring_configs
  • Daily run at 8am SGT
  • Issues include recommended_specialist for routing

4. MCP server

  • Exposed at https://api.adros.ai/mcp
  • Streamable HTTP transport
  • X-User-Token: lyr_... auth
  • Tool categories include patterns, creative generation, campaign actions, memory, monitoring, contracts, workspace, files, and orchestrator tasks

5. REST API

  • Base: https://api.adros.ai/api/v1
  • Auth: Authorization: Bearer lyr_...
  • Live schemas: api.adros.ai/docs (opens in a new tab)
  • Major route groups: orchestrator, workspace, chat, files, campaigns, monitoring, contracts, integrations

6. Webhooks

When configured, Adros POSTs monitor results daily.

Public webhook expectations:

  • HMAC-signed payloads via X-Adros-Signature
  • dedupe_key for consumer-side deduping
  • no retry guarantee

7. Contract layer

Contracts are keyed by (stage, task_type) and define:

  • produce
  • accept
  • recommended_specialist
  • review_gates
  • next_stage

recommended_specialist is a runtime actor ID such as adros_ceo, adros_research, adros_cmo, adros_cd, or adros_deployer.

8. Quality track (audits)

When enabled:

  • scorecards on completed audit tasks
  • guardrails for audit quality thresholds
  • artifact persistence
  • audit trends per project

Confirm support on the environment you are documenting before presenting it as live.

9. Creative pipeline

  • Pattern-anchored generation
  • idempotency protections
  • rate limits
  • output persisted into task artifacts and storage

10. Daily autonomous monitor

Current product framing includes six major checks:

  1. Session health
  2. Creative fatigue
  3. Search term waste
  4. RSA health
  5. Conversion health
  6. Advantage+ audit

11. Common integration pitfalls

  1. Always verify task ownership before mutating task state.
  2. Do not assume file upload alone means task completion.
  3. R2-backed file routes require the target environment to be configured correctly.
  4. Scorecard/guardrail endpoints depend on both code and environment support.
  5. Campaign data depends on connected-account scope.
  6. WebSocket auth uses the token query param.
  7. Webhook receivers should be resilient and idempotent.
  8. recommended_specialist uses runtime actor IDs, not display names.

12. TL;DR

  • Adros = hybrid marketing OS: deterministic rails + managed specialist depth + operator dashboard
  • Runtime actors: adros_ceo, adros_research, adros_cmo, adros_cd, adros_deployer
  • MCP: Streamable HTTP at /mcp
  • REST: all under /api/v1/*
  • Pattern moat: 4,022 patterns
  • Quality track: audit-only and deployment-dependent
  • Artifacts: versioned, reviewable, and traceable across strategy through deploy
Tool CatalogREST — Dashboard & Orchestration