Skip to content
Cload Cloud
Productivity

mercury-mcp

Cheatsheet for the Mercury (Proton) MCP tools. Message agent teammates, manage threads, create tasks, and schedule automations across coordinated agent teams.

What mercury-mcp Does

Mercury MCP (Model Context Protocol) is a coordination toolkit designed for managing multi-agent teams and complex workflows. It enables seamless communication between AI agents, thread management for organized conversations, task creation and assignment, and automation scheduling—all essential capabilities for orchestrating work across distributed agent teams. This skill is built for product designers, operations managers, and power users who leverage AI agents to execute coordinated projects and need centralized control over agent interactions and task pipelines.

How to Install

  1. Access your Mercury/Proton dashboard at teamoffsite.ai
  2. Navigate to Skills or Integrations section
  3. Search for “mercury-mcp” in the skill marketplace
  4. Click “Install” to add the skill to your agent workspace
  5. Authenticate with your Mercury account if prompted
  6. Configure default team settings and notification preferences
  7. Add team members and assign roles (admin, coordinator, agent)
  8. Test the connection by sending a test message between agents
  9. Create your first task to verify full functionality

Use Cases

  • Multi-agent project coordination: Assign tasks to different AI agents handling design, content, and analysis workflows, with Mercury MCP managing dependencies and execution order
  • Customer support escalation: Route support tickets between agent teams, maintain thread history for complex cases, and automate handoffs to human specialists
  • Content production pipelines: Schedule automated workflows where agents research topics, generate drafts, review edits, and schedule publication—all coordinated through Mercury
  • Compliance and approval workflows: Create task chains requiring agent-to-agent reviews and human approvals, with audit trails maintained in Mercury threads
  • Scheduled reporting and monitoring: Set up recurring automations where agents collect data, analyze trends, and deliver reports on set schedules to team members

How It Works

Mercury MCP operates as a coordination layer between Claude instances and other AI models, using the Model Context Protocol standard. When you create a task or message, Mercury generates a structured protocol message that includes context, state, and execution instructions. This message is routed to the appropriate agent teammate or team, which processes it within its own context window while maintaining awareness of the broader conversation thread.

Threads in Mercury function as persistent conversation channels that preserve message history, task status, and context across multiple agent interactions. This prevents information loss and allows agents to reference prior decisions. When you schedule an automation, Mercury’s scheduler triggers task creation at specified intervals, passing the necessary context to agents responsible for execution. The skill provides templated commands for common operations—messaging teammates, creating prioritized tasks, updating thread status—reducing friction in multi-agent workflows.

Under the hood, Mercury maintains a state machine for tasks (pending, in-progress, completed, blocked) and provides webhooks for external system integration. All agent communication is logged within threads, creating an auditable record of decisions and outputs. This architecture allows non-technical users to orchestrate complex agent workflows through simple commands while preserving full transparency into agent actions.

Pros and Cons

Pros:

  • Purpose-built for multi-agent coordination—understands context windows, token limits, and model-specific behaviors
  • Minimal learning curve for non-technical users—template-based task creation and messaging commands
  • Persistent thread history provides full audit trails and eliminates context loss across agent handoffs
  • Flexible automation scheduling with conditional logic—supports retry strategies and escalation workflows
  • Model-agnostic implementation (MCP standard)—works with Claude, GPT-4, Llama, and emerging models without vendor lock-in
  • Integrates seamlessly with existing tools via webhooks and API—connects to CRM, project management, and data platforms

Cons:

  • Requires active Mercury/Proton subscription—not a standalone open-source tool
  • Learning curve for complex automation logic—advanced scheduling and conditional rules may require documentation review
  • Limited customization of task state machines—workflows must fit Mercury’s predefined status and priority model
  • Scaling considerations—unclear pricing tiers for large agent teams or high-volume task throughput
  • Dependency on external service availability—agent coordination halts if Mercury infrastructure experiences outages
  • Context summarization for long threads may lose nuanced details—useful for brevity but risky for compliance-sensitive workflows
  • Claude API Integration: Direct API calls to Claude for building custom agent behaviors within Mercury workflows
  • Zapier/Make Automation: External workflow automation tools that can trigger Mercury tasks and consume thread outputs
  • Slack/Teams Integration: Channel-based notifications and task creation directly from messaging platforms
  • Airtable/Notion Sync: Store and reference task templates, agent instructions, and project data from these platforms within Mercury
  • LangChain Agent Tools: Framework for building more complex agent reasoning and memory management that feeds into Mercury coordination

Alternatives

  • n8n or Make (Zapier alternative): Offers workflow automation but lacks AI agent awareness—designed for API-to-API automation, not agent coordination
  • LangGraph or CrewAI: Both provide multi-agent orchestration but require coding and lack visual task management interfaces—better for developers than product designers
  • Linear or Asana with AI plugins: Traditional project management with basic AI assistants, but not purpose-built for agent-to-agent communication and context passing
Glossary

Key terms

Model Context Protocol (MCP)
An open standard for structured communication between AI models and external tools. It ensures models can understand task context, state, and dependencies across distributed systems.
Thread
A persistent conversation channel in Mercury that maintains message history, task status, and context across multiple agent interactions. Threads prevent information loss and provide audit trails.
Task
A unit of work assigned to an agent or human team member. Tasks have status (pending, in-progress, completed, blocked), priority, dependencies, and execution deadlines.
Agent teammate
An AI model (Claude, GPT-4, etc.) configured to participate in Mercury workflows. Each teammate has assigned roles, task specializations, and communication permissions.
Automation scheduling
The process of configuring Mercury to trigger task creation and agent execution on recurring schedules (hourly, daily, weekly, etc.) without manual intervention.
FAQ

Frequently Asked Questions

What's the difference between Mercury MCP and other team collaboration tools?

Mercury MCP is specifically designed for coordinating AI agents, not just humans. It understands agent context windows, manages task handoffs between models, and automates agent-to-agent communication patterns. Traditional tools like Slack or Asana don't have primitives for agent state management or model-aware task scheduling.

Can I use Mercury MCP with non-Claude agents?

Yes. Mercury MCP implements the Model Context Protocol standard, which is model-agnostic. You can coordinate Claude, GPT-4, Llama, or any MCP-compatible model. The skill handles protocol translation and context formatting for each model type.

How do I set up a task that requires approval before execution?

Create the task with status 'blocked' and assign a human reviewer role. Mercury notifies the reviewer and holds the task until they approve it through the dashboard or via the 'approve-task' command. The task then transitions to 'pending' and gets routed to the executing agent.

What happens if an agent fails to complete a task?

Mercury detects task timeouts (configurable per task, default 24 hours) and routes it to the 'escalation' queue. You can configure escalations to retry with a different agent, notify a human, or create a sub-task. All failures are logged in the thread with error context.

Can I export thread history and audit logs?

Yes. Mercury provides bulk export in CSV, JSON, and PDF formats. All thread messages, task status changes, and agent outputs are exportable. This is essential for compliance workflows requiring audit trails.

How does Mercury MCP handle context window limits across long conversations?

When a thread exceeds an agent's context window, Mercury automatically summarizes older messages and stores full text in a separate archive. The summary is included in the context passed to the agent, preventing token overflow while preserving semantic continuity.

What's the best way to structure tasks for parallel agent execution?

Create sibling tasks (tasks with the same parent) rather than sequential chains. Mercury executes these in parallel. Use the 'wait-for-siblings' command to create a child task that doesn't start until all siblings complete. This pattern prevents bottlenecks in multi-agent pipelines.

Can I integrate Mercury MCP with my existing CRM or project management tool?

Mercury provides webhook endpoints and API access. You can configure webhooks to send task completions to Salesforce, Jira, or Airtable. Conversely, you can trigger Mercury tasks from those systems via API calls, creating bidirectional sync.

More in Productivity

All →
Productivity

outline

Search, read, create, and manage documents in Outline wiki instances (cloud or self-hosted).

sanjay3290
Productivity

google-workspace-skills

Suite of Google Workspace integrations: Gmail, Calendar, Chat, Docs, Sheets, Slides, and Drive with cross-platform OAuth.