Open Source · Apache 2.0
The AI Toolkit for Digital Agencies
The data standardization & context layer for agencies and brands adopting AI
Harness consolidates your organization’s scattered AI workflows and fragmented data, cleaning everything into one unified, AI-ready format — so marketing teams and AI agents can produce reliable work from the same foundation.
Pain Points
When agencies and brands race to adopt AI, the first wall they hit isn’t “the model isn’t smart enough” — it’s that their data simply isn’t ready to be fed to AI. It shows up as three systemic pain points:
Every team and every operator uses different prompts, different tools, and different steps to do the same job — reports, competitive research, KOL discovery. With no standard process, results can’t be reproduced, new hires can’t ramp up, and knowledge never accumulates.
Ad performance lives across Meta, Google, and GA4; social buzz across Threads, Instagram, and Facebook; competitor and KOL data is scattered across even more sources. Fields, units, naming, and collection methods all differ — isolated data silos that must be manually stitched together and cleaned for every single analysis.
Same company, same client — yet this month’s report looks one way and next month’s looks another. One person’s deck never matches a colleague’s layout, colors, fonts, or metric definitions. It looks unprofessional externally and is incomparable internally.
These gaps are the root cause of why AI can’t truly be put to work: feed AI messy, fragmented, inconsistently formatted data and its output will be unstable, untrustworthy, or simply wrong. Tagtoo Agency Harness was open-sourced to fix this at the source.
How It Works
The core value of the toolkit is fully standardizing the two most error-prone, least standardized stages: data collection and data cleaning.
Meta / Google / GA4 ad performance; Threads / IG / Facebook / TikTok / YouTube social and creator data; competitor ads; retail panels; DXP audiences — every source is fetched through Tagtoo-hosted MCP connectors via one single, consistent interface. No more ad-hoc scrapers, per-person API keys, or hand-rolled 403 and pagination handling.
Raw data flows through each workflow’s analysis recipes to be cleaned, deduplicated, and aligned on fields and units, producing structured intermediate formats AI can consume directly: unified analysis JSON, share-of-voice tables, KOL / competitor schemas. Your AI agents receive clean, consistent, trustworthy context — not messy raw responses.
Every deck is rendered through a single brand theme, tagtoo_pptx.js (six byte-identical copies): covers, tables of contents, sections, colors, logos, and fonts are perfectly consistent. A built-in content-aware font selector (Arial for English, PingFang TC for Chinese) eliminates overlapping glyphs and mojibake in mixed-script text.
Not depth in a single feature, but consolidating the daily workflows agencies repeat every day into an out-of-the-box, best-practice-sharing library of standardized skills. New hires are productive immediately, engineers keep full customization flexibility — and the industry finally stops reinventing the wheel every single day.
Workflows
Each workflow is a self-contained module, shipped with a README.md for humans, an AGENTS.md for AI agents, and a conversational SOP instruction.md.
ad-report Reads your clients’ Meta Ads, Google Ads, and GA4 CSVs to auto-generate monthly / quarterly / weekly PPTX reports and daily ad health-check HTML reports — and answers ad questions on the spot. Supports 20+ CSV data types, 10 analysis dimensions, and full KPI computation (ROAS / CPA / CTR / VTR) with period-over-period analysis.
brand-voice-analysis Collects third-party brand mentions across Threads and Instagram, analyzing sentiment, share of voice, and topics. Dual-signal collection (name + hashtag) with deduplication, comparing 1–4 brands at once (your client plus competitors).
competitive-analysis Gathers competitors’ ad placements, creative themes, slots, and social engagement into a “your brand vs. competitors vs. category leader” comparison deck, covering estimated Google (Search / GDN) performance, with the required ad-creative screenshot pages.
retail-analysis Uses Tagtoo retail panel data to benchmark your client’s product and category sales performance against the market.
Retail panel data is proprietary, paid data — request access separately.
dxp A visual HTML interface (Chinese & English) that lets non-technical users browse and filter Tagtoo DXP audience segments (first-party / second-party / retail partner on-site), then trigger downstream audience reports.
The audience catalog UI is deliberately open source; audience membership and activation data are proprietary.
kol-recommendation Searches Instagram and Threads by industry (10 categories) for influencers with sponsored-content experience, detecting sponsorship signals in Chinese and English, ranking by follower count, and producing a recommendation deck per KOL with profile photo, post screenshots, and a “why they fit” rationale.
All social images — profile photos, post thumbnails, ad creatives — are fetched server-side through the connector’s download_media MCP tool and returned as base64, bypassing the 403 blocks that IG / Facebook CDNs impose on client-side fetching, so screenshots embed reliably in every deck.
Architecture
Clear responsibilities and one-way dependencies: Customization → Skills → Tools. Lower layers know nothing about upper layers, fully decoupling workflow logic from data sources — the architectural foundation of “collect data once, reuse it across every client.”
Each client’s dedicated configuration and prompts (not in the public repository).
Cross-client workflows, analysis recipes, and report templates shared across clients.
Private servers wrapped in the MCP protocol: the Tagtoo Social API Connector, gcs-mcp, and Google Drive MCP.
Open Core
All workflow logic, analysis recipes, and report templates are open-sourced under Apache 2.0 — to gain market visibility and let practitioners across Asia easily read, adapt, and contribute. Tagtoo’s real moat is its data and hosted infrastructure: the repository contains no API keys and no free local scraping layer. Running any workflow end-to-end requires access to the Tagtoo Social API Connector (request it from Tagtoo) or standing up an equivalent MCP yourself. Some data sources — the retail panel and advanced DXP segments — are proprietary, paid data.
The core MCP server of the tools layer, deployed on Google Cloud Run. It wraps upstream social data APIs with keys stored server-side, exposing Facebook, Instagram, Threads, TikTok, and YouTube creator queries, plus server-side image fetching via download_media and an /img image proxy endpoint.
Company & Community
The toolkit is maintained by Tagtoo Limited, a Taiwanese company built on digital advertising technology and audience data (DXP), spanning programmatic ad buying and retail / e-commerce data analytics. Contributions via PRs, issues, and discussions are welcome.
As the framework matures, Tagtoo plans “AI manager / AI consultant” services to help clients without engineering resources build and maintain customized versions within the framework — deepening partnerships, with part of the accumulated know-how flowing back into the open-source skills layer.
For business partnerships, connector access, pricing, and adoption consulting, get in touch.
Email [email protected]