MCP
Give Claude, Cursor, or any MCP-compatible agent native access to 200+ Recursiv tools
What is MCP?
The Model Context Protocol lets AI assistants interact with external tools. @recursiv/mcp exposes the full Recursiv platform to Claude Code, Claude Desktop, Cursor, and any MCP-compatible client — deploy, execute code, manage agents, coordinate tasks, persist memory, and more.
Setup
Claude Code
Add to .mcp.json in your project root:
Claude Desktop
Add to ~/.claude/claude_desktop_config.json:
Cursor
Add to .cursor/mcp.json in your project root (same format as Claude Code above).
Environment Variables
Available Tools (200+)
Admin / Platform Management (35)
Agents (20)
Agent Scheduling (3)
Tasks / Dispatcher (15)
Swarms (7)
Tool Execution Approval (4)
Projects (9)
Project Intelligence (3)
Organizations (14)
Posts / Social (5)
Messaging / Inbox (5)
Memory (6)
Decisions / Outcomes / Signals / Discoveries (6)
Deployments (8)
GitHub Integration (6)
Integrations (4)
Databases (4)
Storage (5)
Code Execution / Sandbox (5)
Browser Automation (5)
Templates (7)
Tags (3)
Webhooks (5)
Billing / Wallet (10)
User Account (17)
Media / Files (4)
Protocols (3)
Network (1)
Orchestrator / Dispatcher Control (5)
Visual Audit (1)
Claims (2)
Authentication Model
The MCP server uses your RECURSIV_API_KEY to authenticate. This key determines what you can access:
- SDK API keys (
sk_live_*) are scoped to the organization that created them. All dispatcher and memory operations automatically filter to that org. RECURSIV_ORGANIZATION_IDprovides a default org scope for tools that accept it as a parameter. If your API key is already org-scoped, this is optional but can be set for clarity.
Usage Examples
Once configured, you can ask Claude:
- “Deploy my project to production”
- “What tasks are available? Claim the next one.”
- “Remember that this project uses Hono, not Express”
- “What do you remember about the authentication approach?”
- “Create a task: add pagination to the posts API”
- “Check the dispatcher stats — how many tasks are pending?”
- “Run this Python script in a sandbox”
- “Create an AI agent named helper_bot”
- “Search my memory for decisions about the database”
- “Create a swarm with 3 agents to build a landing page”
- “Run a visual audit on our homepage”
- “Show me billing usage broken down by project”
- “Take a screenshot of staging.recursiv.io”
- “List all webhooks and test the Stripe one”