Forkit
About
Persistent coordination infrastructure for multi-agent AI systems, exposed as a single MCP endpoint.
Details
- Author
- zientesit
- Downloads
- 220
- Categories
- Knowledge Base
Jump to
- Shared task graph with atomic, race-condition-safe claiming
- Automatic dependency graph resolution for task ordering
- Long-poll handoff with ~250ms wake latency
- Execute_code tool runs async JS in a real V8 sandbox
- Collapses multi-step workflows from ~150K tokens to ~1K
- Built on Cloudflare Workers and D1 with usage-based pricing
Setting up with Highlight
This MCP is not yet compatible with Highlight’s one-click setup. However, you can still use it with Highlight by following these steps:
- Download and install Highlight from highlightai.com/download
- Navigate to the plugins tab and select "Add Custom Plugin"
-
Configure the plugin with the settings below
Plugin Name
ForkitCommand (node, npx, python, etc.)Please refer to the README for specific instructions on how to obtain API keys or other required environment variables.
- Enable "Start Automatically" if you want the plugin to start when Highlight launches
From the repository
Integrate Forkit as an MCP server by connecting your MCP client to its endpoint. Agents use the provided tools to create tasks, claim them atomically, resolve dependencies, execute async JavaScript via the execute_code tool, and receive task handoffs via long-poll. No specific configuration keys or installation steps beyond standard MCP client setup are documented.
execute_code
Execute async JavaScript in a real V8 isolate. All DB operations available via codemode.* calls. Returns { result, logs, duration_ms }.
search_api
Return TypeScript type definitions for the execute_code codemode SDK. Call once per session and cache the result.
ready_tasks
Return all pending tasks with no unresolved blockers — the topological ready set. Use as your starting point to find the next task to work on.
claim_task
Atomically claim a pending task for an agent. Race-condition-safe — only one agent wins when two claim the same task simultaneously.
wait_for_task
Block until a ready task is available for the given assignee. Wakes within ~250ms when a task becomes claimable. Use instead of polling ready_tasks.
summarize_session
Generate an AI digest of the current session and persist it to R2 memory as a Markdown file. Call at the end of every work session.
list_executions
Return recent execute_code audit records for this workspace. Each record includes status, duration_ms, and any error.
Claude Desktop / Cursor
Paste into your MCP client config file to install this server.
{
"mcpServers": {
"forkit": {
"forkit": {
"type": "http",
"url": "https://forkit-mcp.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
}McpServers
{
"forkit": {
"type": "http",
"url": "https://forkit-mcp.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}Forkit gives agent swarms shared, persistent infrastructure over a single MCP endpoint. Agents coordinate through a shared task graph — creating tasks, claiming them atomically (race-condition-safe), resolving dependency graphs automatically, and handing off to each other via long-poll with ~250ms wake latency. An execute_code tool runs async JS in a real V8 sandbox, collapsing multi-step workflows from ~150K tokens to ~1K. Built on Cloudflare Workers + D1. First 50 tasks free, then $0.01 USDC/task via x402.
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.
