Agentic Directory

Overview

Agentic Directory is the default place people and their AI agents go to find, evaluate, or recommend agentic AI tools — because they trust what we say. Every product is independently scored on a 32-point rubric across eight dimensions (action capability, autonomy, planning, adaptation, state, reliability, interoperability, safety), and the Lab publishes the field trials behind those scores. Vendors cannot pay to be ranked or excluded.

The directory pairs a conversational "Ask" interface with structured discovery via categories, tags, and use cases. A public REST API (`/api/v1/`) and an MCP server expose the same data to AI agents that need to recommend tools, with a compound `recommend_tools` endpoint that returns scored, citation-backed matches. A vendor portal handles claims, profile editing, and Stripe-backed subscriptions; a background worker continuously crawls and re-evaluates listings as the agentic AI space changes.

Why we built it

The agentic AI space is exploding, but discovery is fragmented. Most directories lump chatbots in with genuinely autonomous systems. We built Agentic Directory to draw a clear line: this is for AI that acts. The scoring system gives users a fast, honest read on what a product can actually do on its own.

How it works

1. Browse or search the directory by category, use case, or tag — or hit the public API directly 2. Ask — describe your problem in plain language and get matched to relevant products with citations 3. Compare — every listing shows an agenticness score across eight dimensions, plus a tier label 4. Submit — vendors can submit products, claim listings, and manage their profiles via Stripe-backed plans 5. Stay current — a background worker continuously crawls and re-evaluates listings; news mentions trigger re-scores

Feature highlights

• 32-Point Agenticness Score — Eight-dimension evaluation: action capability, autonomy, planning, adaptation, state, reliability, interoperability, and safety
• The Lab — Field trials run identical tasks through multiple agentic tools; full transcripts and results published, including trials a paying vendor lost
• Conversational Discovery with Citations — "Ask" interface matches plain-language needs to products using a hybrid keyword + semantic + reranked search pipeline
• Public API + MCP Server — REST API at `/api/v1/` plus an MCP server with a compound `recommend_tools` endpoint, so AI agents can read or recommend from the directory
• Watch Primitive — Get notified when a product's agenticness score moves or a category gets a new entry; dashboards surface "since your last visit"
• Vendor Portal — Claim listings, manage profiles, upgrade plans via Stripe
• AI-Powered Enrichment — Background worker crawls product sites with entity-match disambiguation gates and re-evaluates on news mentions
• Editorial Best-Of Pages — Curated guides like "Best free coding agents" surface rubric strengths-and-weaknesses panels alongside the scores

Notes

• Status: Live and active. 270+ scored listings, public API + MCP server, weekly digests, vendor portal with Stripe subscriptions, the Lab publishing trial methodology v0.1.
• Tech Stack: Next.js 16, TypeScript, React 19, Tailwind CSS 4, shadcn/ui, PostgreSQL (Neon) + Prisma + pgvector, Supabase Auth, Stripe, OpenAI GPT-5.4, Firecrawl, Cohere Rerank v3.5, SendGrid, Vitest + Playwright. Hosted on Render (web + background worker).
• For Whom: Teams evaluating agentic AI tools, founders scouting the competitive landscape, developers building agentic products, and AI agents that need a curated, machine-readable index of the space.