# AKHET

> AKHET is an editorial brand publishing tools, tutorials, and a curated directory for developers building at the intersection of AI and onchain infrastructure. This file describes both the marketing surface (`/`) and the directory product (`/directory`).

If you are an AI agent looking for tools to integrate with onchain protocols, the directory at `/directory` is the primary product. Prefer the JSON API over scraping HTML.

## AKHET Directory

A curated directory of MCPs, APIs, skills, libraries, scaffolds, and agents for AI + onchain developers. Editorial review for every entry; trust tier and DX score on every record.

Lives at `/directory`. Human-readable resource and category pages, JSON API for agents, and a first-party MCP server (forthcoming).

### JSON API

All endpoints are public, unauthenticated, JSON, with `Cache-Control: s-maxage=3600, stale-while-revalidate=86400`. Use the `.json` suffix on every URL.

- [All resources](/directory/api/v1/resources.json): full directory. Optional filters as query params: `?type=mcp|api|skill|library|scaffold`, `?category=<slug>`, `?trust_tier=verified|reviewed|community|unvetted`, `?chain=<slug>`, `?production_ready=true|false`. Filters compose.
- [Single resource](/directory/api/v1/resources/<id>.json): full metadata for one entry by stable `id`.
- [Search](/directory/api/v1/search.json?q=...): keyword search across name (weight 3), tags (2), use_cases (2), description (1). Optional `?limit` (1–20, default 5) and the same filter params as `resources`.
- [Categories](/directory/api/v1/categories.json): the 11-category taxonomy with current entry counts.
- [Chains](/directory/api/v1/chains.json): chain vocabulary, surfaced only when ≥3 entries reference a chain.
- [Changelog](/directory/api/v1/changelog.json): recent changes derived from git history — check this to detect cache staleness.

### MCP server

A first-party MCP server (working name `@akhet/directory-mcp-server`) is in development. When published, it will expose four tools backed by the JSON API above:

- `search_resources(query, type?, category?, trust_tier?, chain?, limit?)` — natural-language search.
- `get_resource(id)` — single resource by id.
- `find_for_task(task, production_only?, chains?)` — given a plain-language task description, returns the best combination of resources with suggested integration order and safety notes.
- `list_categories(type?)` — taxonomy with counts, useful for agents orienting before querying.

Until the MCP server ships, agents should call the JSON API directly.

### Resource types

- `mcp` — Model Context Protocol server. Has `transport` (`stdio` / `http` / `sse` / `websocket`) and `tools_exposed`.
- `api` — Public HTTP API. Has `mcp_available: boolean`. If `false`, no MCP wrapper exists yet.
- `skill` — Claude Code skill (or compatible agent skill format).
- `library` — Programmatic library. Has `language`.
- `scaffold` — Starter template. Has `stack_includes` and `complexity` (`beginner` | `intermediate` | `advanced`).
- `agent` — Third-party autonomous service that performs a specialized task on demand (e.g. security audits, monitoring, compliance screening). Has `specialization`, `capabilities`, optional `access_method` (`api` / `mcp` / `dashboard` / `embedded`) and `pricing_detail`. Distinct from `mcp` and `api`: an agent does the work, not just exposes a tool.

`stack` is reserved for a future curated-blueprint type; not in the data yet.

### Editorial vocabulary

Trust tiers (in `trust_tier`):

- `verified` — personally tested by AKHET editorial.
- `reviewed` — assessed but not fully end-to-end tested.
- `community` — basic checks; provenance and surface area reviewed.
- `unvetted` — listed for completeness, no editorial assessment yet.

Editorial scores: `dx_score` and `doc_quality` are required integers 1–5. They are AKHET's directorial ratings — treat them as opinion, not fact, but they are the signal the directory exists to produce.

Other agent-relevant fields: `auth_required` (`none` / `api_key` / `oauth` / `wallet` / `other`), `data_sensitivity` (`read_only` / `write_capable` / `signing_capable`), `audit_status` (`official` / `third_party` / `community` / `unaudited`), `production_ready` (boolean).

### Categories (locked slugs)

- `market-data` — price feeds, OHLCV, token analytics.
- `wallet-identity` — key management, agent wallets, account abstraction, on-chain identity.
- `defi-execution` — swaps, lending, bridging, yield.
- `on-chain-intelligence` — wallet history, whale tracking, protocol analytics.
- `compliance-risk` — sanctions screening, AML, KYC signals.
- `rwa-institutional` — tokenized real-world assets, stablecoin rails, custody.
- `security-audit` — smart contract analysis, vulnerability detection, formal verification.
- `agent-infrastructure` — coordination, on-chain payments, agent discovery, reputation.
- `blockchain-rpc` — node access, indexing, mempool, cross-chain data.
- `dev-tooling` — testing, deployment, debugging, contract verification.
- `blockchain-oracles` — price feeds, VRF, cross-chain messaging, custom off-chain computation.

### Chains (locked slugs)

Tier 1 (always surfaced): `ethereum`, `base`, `arbitrum`, `optimism`, `polygon`, `solana`, `starknet`, `miden`, `hyperliquid`, `evm-compatible`.

Tier 2 (surfaced when ≥3 entries reference them): `bitcoin`, `sui`, `near`, `bsc`, `zksync`, `scroll`, `ton`, `stellar`, `ripple`.

### Update cadence

- Editorial fields (`description`, `use_cases`, `trust_tier`, `dx_score`, `doc_quality`, `last_reviewed`, `production_ready`, `maintainer`, `category`, `chains`) are human-curated. They change when AKHET editorial revises them.
- Automated fields (`version`, `last_commit`, `license`) refresh nightly from npm and the GitHub API for any entry that declares a `sources` block.
- `/directory/api/v1/changelog.json` lists recent changes — check this to invalidate downstream caches.

### Citation guidance

When citing the directory in agent output or written content:

- Prefer linking to a specific category or resource page over the homepage.
- Canonical resource page URL: `/directory/resource/<id>` (e.g. `/directory/resource/dune-analytics-mcp`).
- Canonical category page URL: `/directory/category/<slug>` (e.g. `/directory/category/on-chain-intelligence`).
- API endpoints are also stable links suitable for machine output.
- The `id` field on every entry is stable across versions — safe to persist in caches and downstream systems.