v0.1 out — tested on mainnet and Fuji

Ask your agent what's up with your validators.

AvaxOps is an open-source MCP server for Avalanche validator ops. Hook it into Claude or Cursor, give it your NodeIDs, and get real answers from the P-Chain — uptime, expirations, balances. Telegram alerts are on the roadmap.

90%ACP-267 uptime floor
5read-only tools
0API keys needed
2networks supported
claude — avax-ops mcp connecting…

Live queries against the public Fuji P-Chain API. Node IDs and address shown are examples; the plain-English summary is formatted client-side, not AI.

Why this exists

Uptime rules got stricter.
The tools didn't get simpler.

Running validators on Avalanche still means a deprecated CLI, a beta alert service, Grafana you host yourself, and whatever scripts you've duct-taped together. It works — it's just a lot to keep track of.

90% or you lose rewards

ACP-267 moved the Primary Network uptime floor from 80% to 90%. You used to have twice the wiggle room. A node that's slowly degrading will hit you in the wallet now.

Too many tools, none of them talk

Avalanche CLI is in maintenance mode. P-Chain stuff goes through Platform CLI, L1s through Builder Console, metrics through Prometheus + Grafana per node. Avalanche Notify is beta — and the docs literally tell you to run your own alerts on top of it.

Your AI can't check your nodes yet

Avalanche's official MCP server only searches docs. Snowpeer and VScout help delegators pick validators — not operators running their own. There's no MCP tool that answers "what's expiring this month?" We're fixing that.

MCP tools

Five tools. All read-only.

They query the public P-Chain API — and your node's local endpoints if you pass a URL. No keys, no signing. Plug into Claude, Cursor, or whatever speaks MCP.

list_validatorsLIVE

Stake, uptime, connectivity, delegators, end date, days_remaining. Pass your NodeIDs, get a table back.

validator_healthLIVE

One validator, full picture. Includes the ACP-267 90% check. Give it your node URL and you also get version, bootstrap status, and health checks.

expiration_reportLIVE

Which validations end in the next N days. Sorted soonest first so you see what's actually urgent.

balance_checkLIVE

P-Chain address balances, or L1 validator fee balances with a rough runway estimate. The fee rate is dynamic — we label it as an estimate.

pingLIVE

Quick sanity check — is the server up, which network, what's the P-Chain height.

diagnose_nodeM3

Not built yet. Plan is log/metrics analysis, top-up estimates, and an L1 config helper. Targeting M3.

Quick start

Should take a few minutes.

No API keys, no account signup. Public P-Chain data just works once you've installed it.

1

Install

Clone the repo, uv pip install -e mcp-server. Needs Python 3.11+.

2

Wire it into your agent

One claude mcp add command. Works with Claude Code, Claude Desktop, or Cursor.

3

Ask something

"Show me validators expiring in the next 30 days" — you'll get live P-Chain data, not a hallucination.

terminal
# 1 — install
git clone https://github.com/EMPangolin/avax-ops.git
cd avax-ops && uv venv && uv pip install -e mcp-server

# 2 — add to your agent
claude mcp add avax-ops -e AVAX_NETWORK=mainnet -- \
  uv --directory $PWD run avax-ops-mcp

# 3 — ask
 "which validators expire in the next 30 days?"
✓ avax-ops OK — network=mainnet, P-Chain height=25181545
Roadmap

What's done, what's next.

10-week build for the Team1 Builder Grant. M1 is shipped. Everything else is in progress.

M1Weeks 1–3 Shipped

MCP server v0.1 — read-only ops

The foundation: query any Avalanche validator straight from your AI agent.

  • P-Chain JSON-RPC client, mainnet + Fuji switchable via AVAX_NETWORK
  • 5 tools — ping, list_validators, expiration_report, validator_health, balance_check
  • ACP-267 90% uptime check built into validator_health
  • pytest suite (mocked RPC) + ruff; verified against live mainnet & Fuji
  • Published to GitHub (MIT) with this live demo site
M2Weeks 4–6 In progress

Monitoring service + Telegram alerts

Stop checking manually — get pinged the moment something needs you.

  • Background poller collecting P-Chain state + node health on a schedule
  • Threshold engine: uptime under 90%, validations expiring, low L1 fee balance, version lag
  • Telegram bot (aiogram): register NodeIDs, real-time alerts, weekly fleet digest
  • Persistence — SQLite for dev, Postgres for prod (SQLAlchemy)
  • Deployed service you can point your own nodes at
M3Weeks 7–8 Planned

AI diagnostics & L1 helpers

Move from "what's wrong" to "here's how to fix it."

  • diagnose_node — root-cause triage from node health + Prometheus metrics
  • estimate_topup — AVAX needed to extend a validation to a target date
  • l1_config_helper — generate and validate L1 genesis / config JSON
  • AvaCloud (Glacier) data integration for richer L1 validator info
M4Weeks 9–10 Planned

Dashboard, docs & launch

A home for operators who'd rather glance than type.

  • Web dashboard — multi-validator and multi-L1 fleet overview
  • One-command MCP install + full setup documentation
  • Tutorials and a demo video for the Team1 community
  • Public beta feedback loop and hardening

Scope stays read-only through v0.x — no transaction signing or submission until a dedicated, security-reviewed milestone. Your keys never touch this tool.

You shouldn't need a spreadsheet to track validators.

MIT licensed, read-only, no accounts. Install it tonight and ask your first question.