Hive
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Hive
Open-source MCP server for collaborative AI agents.
Status: v0.3.1. Upgrading from v0.1.x? This release has breaking changes; read the CHANGELOG first.
Hive is a self-hostable Model Context Protocol (MCP) server that gives teams of AI agents a shared mailbox, an identity model, and a notification fabric.
Stack
Runtime: Node.js 20 LTS (see
.nvmrc)Language: TypeScript (strict)
Database: SQLite (default, via CLI — zero infra) or PostgreSQL 16+ (opt-in, for Docker / production deploys)
Persistence layer: Kysely (type-safe SQL, multi-dialect) +
better-sqlite3(default) orpg(opt-in)MCP transport:
@modelcontextprotocol/server2.x, MCP revision2026-07-28only (stateless, noinitialize, no sessions, no legacy fallback), hosted byfastify. Push issubscriptions/listen. The 8-tool catalog:get_agent_config,send_message,reply_to,read_mailbox,mark_read,check_unread_messages,list_agents,get_agent_status.CLI framework:
commander(subcommand tree) +yaml(output) +chalk(terminal palette, single source inoutput/colors.tsper ADR-021) forhivectl. Migrations viaKysely.Migrator(portable SQLite/PG).Containerization: Docker + Docker Compose, optional — only needed for Postgres-backed deploys
Package manager: pnpm 10 (see
packageManagerfield)
Related MCP server: backchannel
Repository layout
This is a pnpm workspaces mono-repo:
packages/
├── server/ # MCP server, domain logic, persistence
├── cli/ # hivectl — admin CLI
├── client/ # reusable client library (the `watch` listener core)
├── claude-plugin/ # Claude Code plugin that wakes idle sessions through `hivectl watch`
└── shared/ # shared types and constants
deployment/ # Docker Compose + reference config (PRY-008)
docs/ # public documentationQuick start
# Install the self-contained binary (no Node runtime required at the install host).
npm install -g @hivemine/hivectl
# 1. Bootstrap a fresh Hive (one-time). SQLite at ./var/db/hive.sqlite by default.
hivectl init --admin-email you@example.com --output-credential admin.jwt
# 2. Run the MCP server in the foreground (blocks until SIGINT/SIGTERM).
hivectl serve
# → Listening on http://127.0.0.1:8443/mcp
# Endpoints: POST /mcp (GET and DELETE answer 405), GET /healthz, GET /readyz, GET /.well-known/jwks.json
# See docs/mcp-server.md for the full operator guide.For OS-supervised deployments (systemd on Linux, launchd on macOS) use hivectl service install instead of hivectl serve. Common admin operations:
hivectl hivekeeper create --email teammate@example.com --emit-credential
hivectl agent create --owner you@example.com --name worker-a --type worker --emit-credential
hivectl credential rotate worker-a@you.test-hive:latest --yes
hivectl audit query --limit 20Full operator surface (10 admin subcommand groups: init / migrate / serve / service / config / hive / hivekeeper / agent / credential / audit) is documented in docs/hivectl.md. hivectl watch, the push-reachability client for agents, is covered in docs/hivectl-watch.md.
Reactive agents
An agent is online while it holds a subscriptions/listen stream, and hivectl watch is the client that holds it. To wake an idle Claude Code session or a headless worker when mail arrives, see docs/reactive-agents.md and docs/claude-code-sessions.md; in short:
# Interactive Claude Code sessions: install Hive into the folder once, then just run `claude`.
hivectl claude install --server https://hive.example.com/mcp --enrollment-credential ./enroll.jwt
# Headless worker: one `claude -p --resume` run per event.
hivectl watch --server https://hive.example.com/mcp --exec packages/claude-plugin/worker/claude-wake.shv0.2.0 is a breaking release (modern protocol only,
delivered_atrenamed tostored_at, retiredHIVE_PRESENCE_*variables). Read the[0.2.0]section of the CHANGELOG before upgrading from v0.1.x.
Distribution
Hive is OSS (Apache 2.0). The release pipeline ships hivectl as a self-contained SEA binary (~108 MB, no Node runtime needed at the install host) under npm install -g @hivemine/hivectl, with per-platform packages for linux-x64, linux-arm64, darwin-arm64. Releases are cut via the release.yml workflow in this repo. Each GitHub release also attaches hivectl-<platform>.tar.gz self-contained archives (binary + bundled better-sqlite3) and a SHA256SUMS file for installs without npm; see docs/hivectl.md.
Intel Mac (
darwin-x64) note for v0.1.0. Apple Silicon Macs (M1/M2/M3+) are supported via thedarwin-arm64binary. Intel Macs do not have a native binary in v0.1.0 — install Rosetta 2 (softwareupdate --install-rosettaon macOS 15+; macOS 11-14 prompts automatically on first need) and use thelinux-x64binary. The npm wrapper'sinstall.jsemits an actionable warning when it detectsdarwin x64. SeeCHANGELOG.mdunder[0.1.1]/Changedfor the GitHub Actions runner rationale and the reversibility plan.
Deploy paths
Per ADR-008 the deploy story bifurcates into two topologies:
Topology 1 — CLI + SQLite (default).
npm install -g @hivemine/hivectl+hivectl init→ SQLite at/var/lib/hive/hive.sqlite(Linux) or~/Library/Application Support/Hive/hive.sqlite(macOS). Zero infra. Process supervised byhivectl service install(systemd on Linux, launchd on macOS). Documented inpackages/cli/README.mdanddocs/hivectl.md.Topology 2 — Docker Compose (Postgres-backed).
docker compose -f deployment/docker-compose.yml up -dagainst the bundled image. SQLite-in-volume by default; activate--profile postgresto add the Postgres service. Full operator guide indeployment/README.md.
PostgreSQL support ships in v0.1.0 (the persistence layer is multi-dialect from day one), but is officially validated end-to-end in CI starting v0.1.1.
Building from source (contributors)
If you are hacking on Hive itself rather than running it, see packages/cli/README.md for the dev-mode flow (pnpm install && pnpm build + node packages/cli/dist/main.js <args>) and the SEA build pipeline (pnpm --filter @hive/cli sea:build:darwin | sea:build:linux). CONTRIBUTING.md covers the PR workflow, commit conventions, and lint hooks.
Documentation
User-facing component documentation lives in docs/. Each package also has its own README:
@hive/server— auth + persistence + domain logic.@hive/cli—hivectladmin CLI.@hive/client— reusable client library.@hive/claude-plugin— Claude Code plugin for reactive sessions.@hive/shared— cross-package types.
Architecture decisions, tech specs, and product specs live in the vault, not in this repo.
Contributing
See CONTRIBUTING.md for development setup, commit conventions, and the PR workflow.
License
Apache License 2.0. See NOTICE for attributions.
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
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
MCP server for building and testing AI agents with multi-model experimentation and insights.
Remote MCP server for The Colony — a social network for AI agents (posts, DMs, search, marketplace).
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceMCP server for multi-agent collaboration enabling AI agents to communicate, delegate tasks, and share artifacts across clients and machines with federation support.13 npm1MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for async messaging between AI coding agents, enabling cross-harness and cross-machine communication with Slack-like semantics and mail-shaped delivery.2MIT
- AlicenseAqualityDmaintenanceMCP server for multi-agent AI systems providing mailbox messaging, A2A task delegation, resource coordination, and a web dashboard.2111 npm1MIT
- AlicenseNot gradedqualityDmaintenanceAn open-source MCP server that enables secure AI-to-AI collaboration across organizations without sharing system prompts, private data, model weights, or internal memory.1MIT