Tagtoo Agency Harness

The AI toolkit for digital advertising agencies — best practices, ready out of the box.

License: Apache 2.0 Python

Why

Agencies and brands racing to adopt AI keep hitting the same wall — not "the model isn't smart enough," but that their own data isn't ready to feed to AI:

  • Chaotic AI workflows — every operator runs the same task with different prompts, tools, and steps; results aren't reproducible.
  • Fragmented data — ad performance, social voice, competitor and KOL data live in disconnected silos with mismatched fields and units.
  • Inconsistent output — reports differ month to month and person to person in layout, colors, and even metric definitions.

tagtoo-agency-harness is the data-standardization & context layer that fixes this at the source: it unifies collection (all sources through one MCP interface), cleans data into AI-ready formats, and standardizes output (one branded deck theme) — so output is reliable, comparable, and professional no matter who runs it.

Open-core model: the workflow logic, analysis recipes, and report templates in this repo are open source. Tagtoo's moat is the data and the hosted infrastructure — these stay private. Running a workflow that fetches live data requires access to the Tagtoo Social API Connector (and other Tagtoo MCP servers), available on request, or your own equivalent self-hosted MCP. See docs/BUSINESS-MODEL.md.

→ Full overview (EN + 中文): docs/INTRODUCTION.md

Workflows

Directory What it does
ad-report/ Ad-performance monthly / weekly / daily report generator: client CSVs → PPTX (data via Google Drive MCP)
brand-voice-analysis/ Third-party brand mention & share-of-voice analysis across Threads + Instagram → PPTX
competitive-analysis/ Competitor ad-creative strategy intelligence (your brand vs competitors) → PPTX
retail-analysis/ Retail product-sales competitive analysis → PPTX (retail panel data is proprietary — on request)
dxp/ Tagtoo DXP audience selector: HTML UI for choosing audience segments + downstream report
kol-recommendation/ Discover & rank IG/Threads KOLs, detect collaborations → PPTX

Each directory has its own README.md (for humans) and AGENTS.md (for coding agents).

Quick Start

git clone https://github.com/Tagtoo/tagtoo-agency-harness.git
cd tagtoo-agency-harness

Pick a workflow, enter its directory, and follow its README.md. Deck-generating workflows use PptxGenJS:

npm install pptxgenjs

Python-based steps (e.g. ad-report/analyze_data.py) use the standard data stack:

uv pip install pandas numpy python-pptx lxml

Data access: workflows fetch live data through Tagtoo-hosted MCP connectors (Social API, Google Drive, GCS). These require access — request it from Tagtoo ([email protected]) or point the workflow at your own equivalent MCP. No API keys live in this repo.

Documentation

Document Content
docs/ARCHITECTURE.md Three-layer architecture (Customization → Skills → Tools)
docs/BUSINESS-MODEL.md Open-core: what's open vs. proprietary
docs/CONTEXT.md Terminology glossary (chat / plugin / skill / project)
docs/naming-conventions.md Directory / file naming rules
docs/LICENSE-RATIONALE.md Why Apache 2.0
docs/AGENT-ONBOARDING.md Documentation navigation map for coding agents
AGENTS.md / CLAUDE.md One-page agent onboarding at repo root

Contributing

PRs, issues, and discussions are welcome — please read CONTRIBUTING.md first.

Accepted: industry config templates, cross-customer skills/workflows, domain know-how, non-moat scripts, docs, bug fixes. Not accepted: per-customer configs, customer private data, hosted MCP server implementations, and logic tied to Tagtoo proprietary data (retail panel, premium DXP).

License

Apache License 2.0 — see LICENSE and NOTICE. For rationale, see docs/LICENSE-RATIONALE.md.

Maintainers

Tagtoo Limited — partnerships, AI-manager services, onboarding consulting: [email protected]