io.github.hashgraph-online/hashnet MCP Server
io.github.hashgraph-online/hashnet
Universal AI agent discovery and registration across 59,000+ agents in MCP, NANDA, A2A, and OpenRouter ecosystems.
What is the io.github.hashgraph-online/hashnet MCP server?
Hashnet MCP JS is a universal MCP server for discovering, searching, and interacting with AI agents across multiple registries including MCP, NANDA, A2A, and OpenRouter. It provides tools for agent discovery, chat sessions, agent registration, and workflow automation through the HOL Registry Broker ecosystem.
Hashnet enables AI agents to discover and interact with 59,000+ other agents across multiple platforms. It offers search and vector search capabilities, real-time chat with discovered agents, agent registration workflows, and credit management—all accessible through a single MCP interface with support for stdio and HTTP transports.
How to install io.github.hashgraph-online/hashnet
Copy-paste configuration for popular MCP clients.
REGISTRY_BROKER_API_KEYsecretYour Registry Broker API key from https://hol.org
Tools & capabilities
Tools this server exposes to the agent.
hol.stats— Retrieve statistics about available agents and registrieshol.capabilities— Query capabilities across the agent ecosystemhol.search— Search agents by name, description, or metadatahol.vectorSearch— Semantic vector search across agent descriptionshol.resolveUaid— Resolve universal agent identifiershol.chat.readiness— Check agent readiness for chat sessionshol.chat.createSession— Create a new chat session with an agenthol.chat.sendMessage— Send messages within an active chat sessionhol.chat.retry— Retry failed messages in a sessionhol.chat.history— Retrieve chat session historyhol.chat.resume— Resume a previous chat sessionhol.chat.end— Terminate a chat sessionhol.getRegistrationQuote— Get pricing information for agent registrationhol.registerAgent— Register a new agent in the ecosystemhol.waitForRegistrationCompletion— Poll for agent registration completion statusworkflow.discovery— Automated agent discovery workflowsworkflow.delegate— Delegate tasks to discovered agentsworkflow.registration— Automated agent registration workflows
Use cases
- Search and discover AI agents across multiple registries to find specialized tools for specific tasks
- Establish chat sessions with discovered agents to collaborate on complex workflows
- Register new AI agents into the ecosystem and manage their availability
- Automate agent discovery and delegation workflows for multi-agent orchestration
- Query agent capabilities and statistics to understand the broader AI agent landscape
io.github.hashgraph-online/hashnet MCP server FAQ
Hashnet is an MCP server that provides universal discovery, search, and interaction capabilities for 59,000+ AI agents across MCP, NANDA, A2A, and OpenRouter registries. It enables agent-to-agent communication, registration, and workflow automation.
Basic discovery and search tools are available without authentication. Paid tools require a `REGISTRY_BROKER_API_KEY` for access to advanced features like agent registration and premium workflows.
Install via npm: `npm install @hol-org/hashnet-mcp`. Then configure your MCP settings to run `npx @hol-org/hashnet-mcp --stdio` or `--http` depending on your client's transport preference.
Optional: set `REGISTRY_BROKER_API_KEY` for paid features. Optional Hedera or EVM ledger credentials can be configured for blockchain-backed workflows. The server enforces bearer-token auth when binding to non-local hosts.
Hashnet supports stdio (best for local runtimes), Streamable HTTP (recommended), and legacy HTTP+SSE. It includes health check and metrics endpoints for monitoring.
Yes. Hashnet provides workflow tools for automated discovery, task delegation, and agent registration, enabling multi-agent orchestration and coordination across the ecosystem.
README (reference)
Source of truth, from the repository.
Hashnet MCP JS
Universal MCP server for discovery, chat, registration, credits, and workflow automation across the HOL Registry Broker ecosystem.
Ecosystem
| Surface | Link |
|---|---|
| npm package | @hol-org/hashnet-mcp |
| GitHub repo | hashgraph-online/hashnet-mcp-js |
| GitHub releases | Releases |
| Documentation | hol.org/mcp |
| CodeSandbox | Open examples in browser |
Quickstart
pnpm install
cp .env.example .env
pnpm dev:http
Common launch modes:
pnpm dev:stdiopnpm dev:httppnpm dev:http:compatnpx @hol-org/hashnet-mcp --helpnpx @hol-org/hashnet-mcp --stdionpx @hol-org/hashnet-mcp --http --host 127.0.0.1 --port 3333
When installed globally or linked locally, the binary is hashnet-mcp.
Supported Transports
| Transport | Endpoint(s) | Notes |
|---|---|---|
| stdio | process stdin/stdout | best for local agent runtimes |
| Streamable HTTP | /mcp, /mcp/stream | recommended HTTP transport |
| legacy HTTP + SSE | /mcp/sse, /mcp/messages | enabled with FEATURE_LEGACY_SSE=1 |
Runtime utility endpoints:
/healthz/readyz/metrics
Tool Surface
- Discovery:
hol.stats,hol.capabilities,hol.search,hol.vectorSearch,hol.resolveUaid - Chat:
hol.chat.readiness,hol.chat.createSession,hol.chat.sendMessage,hol.chat.retry,hol.chat.history,hol.chat.resume,hol.chat.end - Registration:
hol.getRegistrationQuote,hol.registerAgent,hol.waitForRegistrationCompletion - Workflows:
workflow.discovery,workflow.delegate,workflow.registration
Tool success responses use structured envelopes in structuredContent:
okdatameta
Tool failures return isError: true with structured machine-readable error fields (code, category, retryable).
Configuration
| Variable | Required | Notes |
|---|---|---|
REGISTRY_BROKER_API_URL | no | defaults to https://hol.org/registry/api/v1 |
REGISTRY_BROKER_API_KEY | no | enables paid tools with a static broker API key |
BROKER_REQUEST_TIMEOUT_MS | no | default upstream request timeout is 60000 |
MCP_TRANSPORT | no | http (default) or stdio |
MCP_HOST | no | defaults to 127.0.0.1 |
MCP_PORT | no | defaults to 3333 |
MCP_ALLOWED_ORIGINS | no | comma-separated allow list |
MCP_SERVER_BEARER_TOKEN | no | required when binding to a non-local host (for example 0.0.0.0) |
MCP_SESSION_IDLE_TTL_MS | no | defaults to 900000 |
MCP_SESSION_MAX_COUNT | no | defaults to 250 |
MCP_SESSION_REAP_INTERVAL_MS | no | defaults to 60000 |
LEDGER_ACCOUNT_ID | no | generic ledger identity fallback |
HEDERA_ACCOUNT_ID | no | Hedera account id |
HEDERA_NETWORK | no | e.g. hedera:testnet |
HEDERA_PRIVATE_KEY | no | Hedera private key |
EVM_LEDGER_NETWORK | no | e.g. eip155:1 |
ETH_PK | no | EVM private key |
Security Defaults
- Binds to
127.0.0.1by default. - Validates
Originfor HTTP requests when present. - Enforces bearer-token auth when binding to non-local hosts.
- Reaps idle HTTP sessions and enforces max active sessions.
- Redacts sensitive values in logs.
Development
| Command | Purpose |
|---|---|
pnpm build | compile distributable artifacts |
pnpm start | run compiled server |
pnpm lint | run ESLint |
pnpm typecheck | run TypeScript checks |
pnpm test:run | run Vitest once |
pnpm test:coverage | run Vitest with coverage |
pnpm check:no-stubs | enforce no-stubs contract |
pnpm smoke:http | streamable HTTP smoke test |
pnpm smoke:stdio | stdio smoke test |
Release Notes Automation
GitHub releases are generated automatically from merged pull requests:
- release tags are created during publish (
vX.Y.Z) - GitHub release notes are generated with GitHub's release-note engine
- changelog categories are controlled via
.github/release.yml - canonical docs links are appended to each release
Related MCP servers

HOL Guard
Local-first antivirus for AI agents—detect and block secrets, prompt injection, unsafe commands, and supply-chain risks before execution.

Terraform
Integrate Terraform Registry and HCP Terraform with AI agents for infrastructure automation.
Indexes your Next.js project so an agent finds the file that owns a task in one call.

Giggal.ai
Runs a deep mailbox existence check to verify catch-all, accept-all, and SEG protected emails.
Hiring, SEC, research papers, GitHub & Hacker News as JSON for AI agents. Pay-per-result on Apify.
AI-powered forex and crypto trading analysis with real-time data across 127+ instruments.
