Skip to content

MCP Integrations Overview

Applies to: Dispatch v1.0.0, last verified 2026-04-05

This page explains what MCP is, lists the six tools Dispatch exposes to all connected agents, and links to per-provider setup instructions.

The Model Context Protocol (MCP) is an open standard for exposing tools and resources to AI agents. All four providers supported by Dispatch (Claude Code, Codex, Copilot, and Gemini) connect to MCP servers using the same JSON-RPC 2.0 protocol.

Dispatch runs a built-in MCP server at POST /mcp on the same port as the main dashboard. Any agent with access to that URL can call Dispatch tools directly.

ToolDescription
dispatch_get_stateReturns a summary: epic count, agent count, session count, and which agents are live
dispatch_list_agentsLists all known agents with tool, liveness, current epic, and persona
dispatch_get_epicFull detail for one epic: phases, statuses, plan body, and execution log path
dispatch_update_phase_statusSets a phase status to TODO, IN_PROGRESS, DONE, BLOCKED, ON_HOLD, or CANCELLED
dispatch_emit_activity_eventPushes a typed event into the live activity feed
dispatch_get_session_contextReturns transcript path and session metadata for a given session ID

In Claude Code, these appear as mcp__dispatch__<tool_name>.

GET http://localhost:<port>/mcp

Returns server name, protocol version, and the full tool list. No auth required. Replace <port> with the port shown in the terminal when the server starts.

By default (localhost dev), all requests are allowed through. To require a bearer token, set the DISPATCH_MCP_TOKEN environment variable before starting the server. Then pass Authorization: Bearer <token> in requests. Auth is checked with a constant-time comparison to prevent timing attacks.

Each provider connects to the same MCP endpoint but configures it differently:

  • Claude Code: Dispatch auto-injects .mcp.json into every registered project. Zero manual steps.
  • Codex: Dispatch auto-injects .codex/config.toml when a .codex/ directory is detected. Manual snippet also available.
  • Copilot: Requires a manual snippet in VS Code settings. Auto-inject is not possible for VS Code.
  • Gemini: Connect via the Gemini SDK’s built-in MCP support, or import the OpenAPI spec into Vertex AI Agent Builder.