hydronode-mcp
This server has a remotely hosted endpoint. Open its connector listing for connection details, authentication requirements, and setup instructions:
Ask Claude, ChatGPT, Cursor, VS Code or Codex about your HydroNode sensors in plain language, and let them switch outputs, send commands, tick off messages or draft automations with the rights you grant.
Connect in 30 seconds
Hosted endpoint: https://hydronode.tech/mcp
Client | Setup |
Cursor | Add to Cursor or add |
VS Code | Install in VS Code or add |
Claude.ai / Claude Desktop | Settings → Connectors → Add custom connector → |
Claude Code |
|
ChatGPT | Settings → Connectors → Add MCP connector → |
Codex | Add |
Claude plugin (connector + skills) |
|
Clients without remote MCP | run the bridge |
On first use your browser opens HydroNode's OAuth 2.1 consent page so you can choose which projects
the assistant may see and whether it gets read-only access (hydronode:read) or edit access
(hydronode:write, hydronode:automations). Manage or revoke connected apps anytime under
https://hydronode.tech/account/security#connected-apps.
Related MCP server: AetherEdge
Architecture
Claude / ChatGPT ──HTTPS──► <host>/mcp ──Traefik──► hnmcp (this repo: stateless, no DB, no secrets)
│ Agent API with the user's own token
▼
hnbackend (http://hnbackend:8080, internal only)Hosted only. Remote MCP over Streamable HTTP at
https://<host>/mcp, stateless, JSON answers. No local mode.OAuth 2.1 only. Clients discover the HydroNode authorization server via RFC 9728 metadata, register themselves (DCR or Client ID Metadata Document) and the user picks scopes and projects on HydroNode's consent page.
Condensed answers. History comes as ≤ 48 buckets with min/avg/max, lists as ≤ 20 items. Never raw dumps.
Isolated.
hnmcphas no database access and no server secrets. It forwards the caller's agent token (hoa_…) to the backend's Agent API over the internal network. Agent tokens work nowhere else: Traefik blocks them on every public route except/mcp, and the backend only accepts them without proxy headers.
Contracts: hydronode-backend/docs/api/OAUTH.md, hydronode-backend/docs/api/AGENT_API.md.
Tools (19)
Tool | Kind | What it does |
| read | Projects, device counts by status, open anomalies, unread messages, what needs attention |
| read | Sensors, data points, stations, projects by name, id or DevEUI |
| read | One sensor with status, battery, signal, firmware and every data point |
| read | Min/max with time, average, trend, gaps, ≤ 48 buckets |
| read | When a value was below/above a threshold |
| read | Up to 4 data points, or one point against an earlier span ( |
| read | Detected anomalies and their explanation |
| read | Inbox |
| read | Only devices with problems |
| read | Latest device requests with rejection reasons |
| read | Public stations by name or location |
| read | Automations, one with recent runs |
| write | Switch or set a writable data point (destructive hint) |
| write | Queue a typed device command (destructive hint) |
| write | Record a manual measurement |
| write | Read, unread, archive, restore, read all |
| write | Automation on/off |
| automations | Dry run by default; no webhooks ever |
Write tools are always listed. Without the scope they answer with a clear text and a link to Connected apps,
and the HTTP answer is 403 with WWW-Authenticate: Bearer error="insufficient_scope", scope="…" so clients
that support step-up reopen the consent (switch HNMCP_STEPUP_HTTP).
Prompts: daily_report, why_offline, do_i_need_to_water, compare_weeks.
Endpoints
Path | |
| MCP (Streamable HTTP, stateless). Needs |
| RFC 9728 metadata |
| health (ready also checks the backend) |
Configuration
Env | Default | |
|
| listen address |
| required | internal backend, e.g. |
| required | public origin, e.g. |
|
| OAuth issuer |
|
| browser origins allowed on |
|
| requests per token per minute (the backend has its own limits) |
|
| answer missing scopes with HTTP 403 step-up |
|
|
Security
Accepts only
hoa_agent tokens; PATs, refresh tokens and JWTs are refused without a backend call.Token check via
GET /api/agent/v1/me, cached 30 s by SHA-256 hash (rejected tokens 10 s, one lookup for concurrent requests); revocation applies within 30 s here and immediately in the backend.Only Agent API paths can be called, ids are validated, redirects are never followed.
Origin allowlist, 64 KiB body limit, 20 s per request, per-token rate limit, at most 300 rejected tokens per minute and address (above that only already verified tokens pass).
Logs carry tool name, status, duration and a 12-character token hash. Never tokens, arguments or answers.
distroless nonroot image, read-only filesystem, no capabilities.
Develop
go test -race ./...
go run honnef.co/go/tools/cmd/staticcheck@v0.8.2 ./...
go run golang.org/x/vuln/cmd/govulncheck@v1.9.0 ./...
# against a local backend (mvn quarkus:dev on :8080, OAuth + Agent API on in %dev)
HNMCP_LISTEN=:8091 HNMCP_BACKEND_URL=http://localhost:8080 \
HNMCP_PUBLIC_URL=http://localhost:8091 HNMCP_AUTH_SERVER=http://localhost:8080 \
HNMCP_ALLOWED_ORIGINS=http://localhost:6274 go run ./cmd/hnmcp
npx @modelcontextprotocol/inspector # connect to http://localhost:8091/mcpRelease
Branch | Workflow | Image |
any, PR |
| none (vet, gofmt, staticcheck, race tests, govulncheck) |
|
|
|
|
|
|
Listed in the MCP Registry as tech.hydronode/hydronode (server.json,
docs/REGISTRY.md). Claude Connectors Directory texts: docs/CLAUDE_DIRECTORY.md.
License
MIT, see LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
Build, validate, deploy — HTTP APIs, cron jobs, webhooks and MCP tools — from your AI client.
Phone, SMS and email for AI agents. One remote MCP endpoint, OAuth login, zero install.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceConnects AI assistants to physical devices via MCP, enabling LLMs to reason about sensor data, perform cross-device correlation, and interpret anomalies.4Apache 2.0
- AlicenseBqualityCmaintenanceExposes a deterministic IoT edge runtime — Modbus/CAN/J1939 acquisition, local rules, alarms, and history — to AI assistants as typed, governed MCP tools. Reads are safe by default; device control stays deny-by-default, explicitly confirmed, and audited.233Apache 2.0

ZopDev MCPofficial
AlicenseNot gradedqualityBmaintenanceEnables AI assistants to query and, when scoped writes are enabled, manage cloud estate data—inventory, costs, recommendations, schedules, governance, and deployments—across AWS, Azure, and GCP through a hosted remote MCP endpoint.Apache 2.0- AlicenseNot gradedqualityAmaintenanceEnables AI clients such as Claude, Cursor, VS Code, and Windsurf to list and manage workspaces, projects, tasks, AI agents, and automations through a hosted OAuth-secured endpoint or a local stdio package. Also generates MCP tools from any OpenAPI spec.2,530 npmMIT