Batchit

Batchit is an open-source MCP server that enables batching of multiple Model Context Protocol (MCP) tool calls into a single request, reducing token usage and network overhead for AI agents.

Features

• Batch Execution: Aggregate multiple MCP tool calls into a single batch_execute request.
• Parallel Execution: Run multiple operations simultaneously, with configurable concurrency via maxConcurrent option.
• Error Handling: Option to stop remaining operations if one fails (stopOnError).
• Timeout Control: Set timeout limits for the entire batch using timeoutMs.
• Connection Caching: Reuses connections to downstream MCP servers for efficiency.
• Single Target Server Per Batch: Each batch call is directed to one MCP server.
• Consolidated Results: All operation responses are returned together at the end of the batch.
• Installation via Git: Clone and run with TypeScript/Node.js.
• Compatible with Cursor: Can be added globally or per project in Cursor environments.

Limitations

• No data passing between operations within a batch; dependent operations require separate batch calls.
• Each batch references only a single target MCP server.
• All results are returned together at the end of the batch.

Usage Examples

• Reading multiple files in one batch
• Creating directories and writing multiple files based on prior operations (requires phased requests)

Category

mcp-middleware-orchestration

Tags

batching, orchestration, ai-agent, open-source

Pricing

No pricing information is provided; Batchit is open-source.

Source

GitHub repository