veegres/ivory v2.1.0 on GitHub

🤖 MCP Server for AI Agents

Ivory now exposes its features over the Model Context Protocol, so an AI agent (Claude, Cursor, Codex, VS Code, …) can read a cluster's health and operate it in plain language — "why is replica 2 lagging?", "what's blocking my queries?", "switch the leader to node 3".

Streamable HTTP endpoint at /api/mcp, served on the same address and port as the web UI
Same permissions as the user: every tool is gated by the signed-in user's permissions, exactly like the UI
Everything Ivory knows becomes one conversation: node roles, replication lag, pending restarts, drift warnings, VM metrics, container logs and saved diagnostic queries
New Settings → Integrations → MCP Server screen: shows the endpoint, generates a token and gives a ready-to-copy config for Claude Code, Cursor, VS Code, Codex, Antigravity, IntelliJ IDEA and Claude Desktop
New check tools: check endpoints for clusters, keepers, platforms and database connections, so an agent (or you) can verify connectivity step by step
Cross-origin protection on the MCP handler

See the MCP server docs for setup, the tool list and permissions.

🔐 Authentication

Session tokens for agents: POST /auth/token mints a bearer token with a lifetime you pick (1h / 6h / 1d / 15d / 30d, never unlimited). It writes no cookie, can't mint another token, and is re-checked against the user store
Login sessions renew themselves: an active browser stays signed in, and an idle one signs out after 30–60 minutes
Common request validation: every HTTP and MCP request is validated through the same binding rules, so both report errors the same way

🚀 Cluster Deployment

Cluster deploy and node deploy behave the same way, sharing one flow and one dialog
Per-step deploy status: each node reports its deploy and initialize steps as done, failed or skipped, so a partial deploy is obvious
Container inspect for deployed nodes

🩺 Keeper Health & Monitoring

More warnings from keeper adapters: engines now report more of what theyto the node they actually happened on
Warnings merged for the same node across responses
Lag display fixed across engines. An unreachable node now reports unknown lag (-1) instead of 0
Better Postgres lag check
Fewer requests to etcd
Fixed ClickHouse plugin delay
Fixed the keeper request sharing the same timeout as the cluster request
Built-in queries are updated on restart unless the user changed them

🎨 UI & Branding

New butterfly logo and favicon
Header fixed to fit mobile screens
Improved docs and README

🧹 Maintenance

Router files split by transport (HttpRouter / McpRouter), with transport types named Router***
Go warnings fixed
Dependency bumps

⚠️ Compatibility

No fresh installation required. v2.1.0 is a drop-in upgrade from v2.0.0: storage and backup format (ivory.v2.bak) are unchanged.

After upgrading:

New permissions view.cluster.check, view.node.keeper.check, view.node.py.db.check are added. Superusers get them automatically; grant them to otherusers in Settings → Permissions if needed
Browser sessions now expire after inactivity rather than at a fixed time

What's Changed

add mcp server by @anselvo in #756
bug fixes and other improvements by @anselvo in #760

Full Changelog: v2.0.0...v2.1.0
Changed: 246 files (+9,489, -1,939 lines)