Shift Server

---

Tools

Two tools bracketing a working session — one to announce it, one to end it:

Tool Name	Description
shift_check_in	Register or update a worker session. Returns a worker ID, the coordination protocol, and the active peers.
shift_check_out	End a working session and remove it from the active worker list.

shift_check_in

Called at the start of every working session, and again whenever the scope changes.

• Takes a gist of the current work plus optional files the agent expects to modify
• Returns a 6-character worker ID, the coordination protocol, and the full active-workers table
• Pass an existing workerId to update the session — patch semantics, so omitted fields and the original check-in timestamp are preserved
• An unknown workerId fails with reason: "unknown_worker" and embeds the active-workers table so the caller can self-identify or start fresh

---

shift_check_out

Ends a working session.

• Takes a workerId and an optional one-sentence summary
• Idempotent — an unknown or already-checked-out ID succeeds silently

Resources

Type	Name	Description
Resource	shift://status	All currently active workers with their gists, declared files, and check-in timestamps.

The same roster is returned inline by shift_check_in, so tool-only clients see it without reading the resource. Subscribers to shift://status receive notifications/resources/updated on every check-in, session update, and check-out.

Features

Built on @cyanheads/mcp-ts-core:

• Declarative tool and resource definitions — one file per primitive, framework handles registration and validation
• Unified error handling — handlers throw, framework catches, classifies, and formats
• Structured logging with request-scoped context and optional OpenTelemetry tracing
• STDIO and Streamable HTTP transports, serving MCP protocol revisions 2025 and 2026-07-28

Coordination-specific:

• In-memory worker session store — no database, no filesystem writes, cleared on restart
• The coordination protocol ships in every check-in response, so ground rules reach the agent without client-side configuration
• The active-workers table rides along with every check-in for situational awareness
• Patch semantics on session updates — only the fields provided change

Agent-friendly output:

• Both client surfaces carry the same data — structuredContent from the output schema, markdown from format()
• shift_check_in declares a typed error contract, so an unknown worker ID arrives with data.reason and a recovery hint
• Failure responses embed the current roster, so an agent recovers in the same turn instead of calling again to orient

Getting started

Add the following to your MCP client configuration file:

{
  "mcpServers": {
    "shift-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/shift-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "shift-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/shift-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "shift-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "MCP_TRANSPORT_TYPE=stdio",
        "ghcr.io/cyanheads/shift-mcp-server:latest"
      ]
    }
  }
}

Every agent sharing a codebase must reach the same server process for the roster to be shared. Over stdio each client spawns its own process, so point concurrent agents at one Streamable HTTP instance instead:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.3.0 or higher (or Node.js v24+).
• No API keys, accounts, or external services.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/shift-mcp-server.git

2. Navigate into the directory:

cd shift-mcp-server

3. Install dependencies:

bun install

4. Configure environment:

cp .env.example .env
# edit .env if you need to override a framework default

Configuration

No server-specific environment variables. Framework defaults worth knowing:

Variable	Description	Default
MCP_TRANSPORT_TYPE	Transport: stdio or http.	stdio
MCP_HTTP_PORT	Port for the HTTP server.	3010
MCP_HTTP_HOST	Hostname for the HTTP server.	127.0.0.1
MCP_SESSION_MODE	HTTP session handling: auto, stateful, or stateless. auto resolves to stateful; this server runs stateless, since no handler needs a session.	auto
MCP_AUTH_MODE	Auth mode: none, jwt, or oauth.	none
MCP_LOG_LEVEL	Log level (RFC 5424).	info
LOGS_DIR	Directory for log files (Node.js only).	<project-root>/logs
OTEL_ENABLED	Enable OpenTelemetry instrumentation.	false

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run:

  bun run rebuild
  bun run start:stdio
  # or
  bun run start:http
  

• Run checks and tests:

  bun run devcheck   # Lint, format, typecheck, security, packaging
  bun run test       # Vitest suites: unit, smoke, integration, fuzz
  bun run lint:mcp   # Validate MCP definitions against spec
  

Docker

docker build -t shift-mcp-server .
docker run --rm -p 3010:3010 shift-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/shift-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Directory	Purpose
src/index.ts	createApp() entry point — registers the tools and the resource.
src/mcp-server/tools	Tool definitions (check-in.tool.ts, check-out.tool.ts).
src/mcp-server/resources	Resource definitions (status.resource.ts).
src/services/worker-store	In-memory worker session store and table formatting.
tests/	Unit, smoke, integration, and fuzz suites mirroring src/.

Development guide

See CLAUDE.md/AGENTS.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for request-scoped logging, ctx.state for tenant-scoped storage
• Register new tools and resources in src/index.ts
• format() must render every field in the output schema — both client surfaces carry the same data

License

Apache-2.0 — see LICENSE for details.