MCP Server

depends.cc works as an MCP (Model Context Protocol) server, letting AI agents in Claude Code, Cursor, Windsurf, and other tools query and update dependency state directly.

Setup

The MCP server runs in two modes:

Local mode — connects to a local depends serve instance. No signup needed, no limits.
Remote mode — connects to depends.cc using your token. Requires depends signup first.

Mode is auto-detected: if no token is configured, it uses local mode. If ~/.depends/config.yml has a token, it uses remote mode.

Claude Code

Add to your project's .mcp.json:

// Local mode (auto-detected when no token is set)
{
  "mcpServers": {
    "depends": {
      "command": "depends",
      "args": ["mcp"]
    }
  }
}

// Remote mode (explicit)
{
  "mcpServers": {
    "depends": {
      "command": "depends",
      "args": ["mcp"],
      "env": {
        "DEPENDS_TOKEN": "dep_..."
      }
    }
  }
}

Or add globally via:

# Local
claude mcp add depends -- depends mcp

# Remote
claude mcp add depends -e DEPENDS_TOKEN=dep_... -- depends mcp

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "depends": {
      "command": "depends",
      "args": ["mcp"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "depends": {
      "command": "depends",
      "args": ["mcp"]
    }
  }
}

Available tools

Once configured, the AI agent has access to these tools:

depends_status: Get the state of all nodes or a specific node. Returns JSON with state, effective state, reason, and solution.
depends_set: Set a node's state to green, yellow, or red. Supports reason and solution fields.
depends_graph: Get the full dependency graph — nodes, edges, and effective states.
depends_events: List recent state change events for a namespace or specific node.
depends_push: Upload a depends.yml file to sync the dependency graph.
depends_pull: Download the current graph as YAML.

Example usage

Once the MCP server is configured, AI agents can interact with your dependency graph naturally:

Checking status before deploying

"Check if all dependencies for the api-server are green before we deploy."

The agent calls depends_status and inspects effective states, reporting any upstream issues.

Updating state from CI

"The database migration failed — mark the database as red with the reason."

The agent calls depends_set with state red, a reason, and a suggested solution.

Investigating incidents

"What changed in the last hour? Something is broken."

The agent calls depends_events to find recent state changes and traces the root cause through the dependency graph.

AI onboarding

When an AI agent encounters depends.cc for the first time, these resources help it get up to speed:

llms.txt

A plain-text overview at llms.txt — designed for AI agents to fetch in a single request. Covers key concepts, CLI commands, and the full API surface.

JSON docs

Request /docs with Accept: application/json for structured, machine-readable API documentation including every endpoint, method, auth requirement, and parameter schema.

curl -H "Accept: application/json" https://depends.cc/docs

Status as JSON

Use depends status --json or the API directly to get machine-readable state for all nodes in a namespace.

depends status --json
depends status myproject/database --json

Local vs remote

Local

Run depends serve and the MCP server connects to it automatically — no signup, no token, no limits.

# Start local server
depends serve

# MCP auto-detects local mode — nothing else to configure

Remote

Run depends signup once. The token is saved to ~/.depends/config.yml and the MCP server switches to remote mode automatically.

# One-time setup
depends signup

# MCP now connects to depends.cc using your saved token