agent-link-mcp

# agent-link-mcp

MCP server for bidirectional AI agent collaboration. Lets any MCP-compatible agent spawn and communicate with other AI coding agent CLIs.

## Tools

### spawn_agent
Spawn an AI agent CLI (claude, codex, gemini, aider, or custom) to work on a task. Supports model selection, thinking depth, git diff context, auto-retry with escalation. Default timeout: 1 hour.

Parameters:
- agent (required): Agent name ("claude", "codex", "gemini", "aider")
- task (required): Task description
- context: Optional { files, error, intent, diff }. diff: true includes git diff output, "staged" for staged only.
- cwd: Working directory for the agent process
- model: Model to use (e.g. "o3", "gpt-5.4", "claude-sonnet-4", "gemini-2.5-pro"). Passed via --model flag.
- thinking: Thinking/reasoning depth ("low", "medium", "high", "max"). Claude uses --effort, Codex uses -c reasoning_effort, Aider uses --reasoning-effort.
- retry: Auto-retry on failure (up to 3 attempts). Default: false.
- escalate: On retry, automatically increase thinking level. Requires retry: true. Default: false.
- timeoutMs: Timeout in ms. Default: 3600000 (1 hour).

### spawn_agents
Run multiple agents in parallel. Each agent runs independently and all results are returned together. Accepts array of agent specs with individual model/thinking/context settings. Great for parallel code reviews, getting multiple opinions, or distributing subtasks.

### reply
Continue a conversation with a spawned agent that asked a question. Send your answer and get back the next response.

### kill_agent
Abort a running agent session by ID.

### list_agents
List all configured agent CLIs and their availability on the system. Auto-detects installed CLIs.

### get_status
Get status of all active agent sessions including conversation history and process info.

## Prompts

### collaborate
Generate a prompt for asking another agent for help with any task.

### debug-with-agent
Ask another agent to help debug an error with file context.

### code-review
Request a code review from another agent on specified files.

## Resources

### agent-link://agents
JSON list of all configured agents and their availability.

### agent-link://config
Current agent-link configuration and profiles.

## When to Use

- You've tried to fix a bug twice and keep failing — use retry + escalate, or ask another agent
- You want parallel code reviews from multiple models — use spawn_agents
- You want to include current git changes as context — use context.diff
- You need to leverage different model strengths (Claude for planning, Codex for execution, Gemini for research)
- You want to control thinking depth — use the thinking parameter
- You want auto-retry with increasing reasoning effort — use retry + escalate