claude-agent-dispatch
Multi-provider agent dispatch plugin for Claude Code.
Route tasks to Anthropic, Ollama (local + cloud), OpenAI, Groq, or any compatible provider — with parallel execution, tier spoofing, live status, and cost tracking.
Highlights
- Multi-provider — Anthropic, Ollama, OpenAI, Groq through one MCP interface
- True parallel dispatch — multiple agents run concurrently in a single message
- Tier spoofing — Ollama models adopt Claude identity via model aliasing
- Built-in agents — general, explore, reviewer, implementer, or bring your own
.md
- Interactive sessions — split terminal pane for human-in-the-loop (Windows Terminal, tmux)
- Live status line — real-time token counts, cost, elapsed time, hang detection
- Audit trail — JSONL log of every dispatch with cost and token accounting
- Agent-intercept hook — optionally redirect all native
Agent tool calls through dispatch
- Interactive configurator — menu-driven model/provider management with Ollama auto-scan
Requirements
| Requirement | Notes |
|---|
| Claude Code | CLI — desktop app, web app, or IDE extension |
| Python 3.10+ | Runs the MCP server and tools |
mcp package | pip install mcp |
| Ollama | Optional — for local/cloud model routing |
| OpenCode | Optional — for OpenAI/Groq backends |
Installation
From Claude Code (recommended)
Option 1 — Direct install:
/plugin install wezz-solutions/claude-agent-dispatch
Option 2 — Add as marketplace, then browse:
/plugin marketplace add wezz-solutions/claude-agent-dispatch
Then open /plugin → Discover tab → install from there.
Install scopes:
/plugin install wezz-solutions/claude-agent-dispatch --scope project # shared via git
/plugin install wezz-solutions/claude-agent-dispatch --scope user # all your projects
/plugin install wezz-solutions/claude-agent-dispatch --scope local # this project only, gitignored
The plugin automatically:
- Registers the MCP server (6 dispatch tools)
- Creates default
dispatch.json on first session
- Makes
/dispatch and /dispatch-configure skills available
- Loads built-in agents (general, explore, reviewer, implementer)
Uninstall
/plugin uninstall claude-agent-dispatch
Manual install (alternative)
If you prefer not to use the plugin system:
git clone https://github.com/wezz-solutions/claude-agent-dispatch.git
python claude-agent-dispatch/scripts/install.py # project-level
python claude-agent-dispatch/scripts/install.py --user # user-level
python claude-agent-dispatch/scripts/install.py --enforce # + Agent-intercept hook
Restart Claude Code after manual install.
Quick Start
Once installed and Claude Code restarted:
# Dispatch a task
mcp__dispatch__run(
query="Fix auth bug",
prompt="Fix the JWT validation bug in auth.py...",
model="sonnet"
)
# Parallel dispatch — all run concurrently
mcp__dispatch__run(query="Fix auth", prompt="...", agent="implementer", model="sonnet")
mcp__dispatch__run(query="Add tests", prompt="...", agent="implementer", model="haiku")
mcp__dispatch__run(query="Review API", prompt="...", agent="reviewer", model="haiku")
# Use Ollama model
mcp__dispatch__run(query="Quick task", prompt="...", model="ollama/kimi-k2.6:cloud")
# Interactive session (opens split terminal pane)
mcp__dispatch__interactive(query="Debug", prompt="Help debug the auth flow", model="sonnet")
# Check status and configuration
mcp__dispatch__status()
mcp__dispatch__config()
Skills (slash commands)
The plugin registers two Claude Code skills:
| Command | Purpose |
|---|
/dispatch | Quick reference — tools, models, agents, parallel usage |
/dispatch-configure | Launch the interactive configurator |
MCP Tools
| Tool | Purpose | Blocking |
|---|
run | Dispatch headless agent | Yes — returns result when done |
interactive | Open agent in terminal pane | No — returns immediately |
status | List active/recent dispatches | No |
result | Get completed dispatch result | Polls up to 60s |
cancel | Terminate a running dispatch | No |
config | Show current configuration | No |
Parameters for run and interactive
| Parameter | Required | Description |
|---|
query | Yes | Short label, max 20 chars (e.g. "Fix auth bug") |
prompt | Yes | Full task description |
agent | No | Agent type: general, explore, reviewer, implementer, raw, or path to .md file |
model | No | Model alias: "sonnet", "haiku", "ollama/kimi-k2.6:cloud", "openai/gpt-4.1" |
Configuration
