Overview

Neuronto is a federated Agentic Resource Discovery (ARD) registry — a search index that answers the question “what tool can do this task?” on behalf of AI agents. Instead of hard-coding integrations one by one, an agent sends a single query and Neuronto searches its own catalogue plus every other public ARD registry at once, fusing the rankings into one answer. At the time of writing it indexes more than 124,000 agentic resources from around 67,000 publishers, covering MCP servers, A2A agents, skills, OpenAPI-described REST APIs and plugins. ARD is an open specification (published June 2026) and Neuronto is released as open source under the Apache-2.0 licence.

Key Features

• Federated search: one query fans out across every public ARD registry concurrently and merges the results — selectable as auto, referrals or none
• Verified tool index: handshakes with each indexed MCP endpoint and reads its own tools/list, so entries carry real tool names and input schemas rather than publisher prose (32,000+ verified tools across 2,200+ servers)
• Hybrid retrieval: sparse BM25 lexical search fused with dense semantic vectors using reciprocal rank fusion
• Verified liveness: indexed endpoints are probed and non-responding ones demoted in ranking rather than deleted
• Multi-protocol access: REST (POST /search, /explore, /tools, GET /agents), MCP (POST /mcp) and A2A (POST /a2a) — installable as an MCP server in one command
• Open API, no key required: the whole index is queryable with no signup; a verified-domain key only raises rate allowances
• Manifest generation from evidence: builds an ARD manifest for a domain from resources that actually answered, recording what proved each entry — nothing is inferred
• Payment terms up front: records each resource’s cost (pay:protocol, price, currency, network) so an agent knows the price before calling
• Private entries: a DNS-verified domain can register internal services in separate storage, returned alongside public results and clearly labelled

Why Use It

An agent can only use capabilities that someone wired in beforehand, and every tool description competes for space in the context window — a model that does not scale to thousands of tools. ARD moves that selection problem out of the context window and into a search service, the same shift the web made from curated directories to search engines. Neuronto is the discovery layer rather than a runtime: it tells an agent what exists, and the agent connects using the resource’s own protocol (MCP, A2A, OpenAPI). Because it aggregates the official MCP Registry, the major community aggregators and its own crawl of the well-known discovery paths, it reaches publishers that no single source covers alone. The API is open and free to query, and a self-hosted Neuronto charges nothing at all.

Use Cases

• Agent builders who want to stop hard-coding integrations and resolve capabilities at runtime
• API and tool vendors who want to become discoverable by publishing one manifest at /.well-known/ard.json instead of applying to a curated marketplace
• Platform teams running discovery over internal services so in-house agents find them the same way they find public ones
• Researchers auditing which MCP servers, skills and agents rank for a given capability across every public registry

Platforms

Hosted registry with a public REST, MCP and A2A API — search from any HTTP client (curl -s https://neuronto.com/search -H 'content-type: application/json' -d '{"query":{"text":"scrape a website"}}') or as an MCP server (claude mcp add --transport http neuronto https://neuronto.com/mcp). Open source and self-hostable for private deployments.

Licence

Open Source — Apache License 2.0. The hosted service is free to query without a key; heavy calls beyond the free allowances are metered with prepaid Neuronto credits (1 credit = $0.001, paid in USDC over x402), and a self-hosted instance charges nothing.

Website

https://neuronto.com

Alternatives

• LocalAI — self-hosted OpenAI-compatible API for local AI models (MIT)
• Open WebUI — self-hosted interface for local and remote AI models (open source)
• Ollama — run and serve large language models locally (MIT)
• The official MCP Registry (github.com/modelcontextprotocol/registry) — open-source registry for MCP servers
• Browse all self-hosted tools — the full TWiF directory