Sefaria Mcp

Sefaria MCP Server

A modern MCP (Model Context Protocol) server for accessing the Jewish library via the Sefaria API.

What does this server do?

This server exposes the Sefaria Jewish library as a set of 15 MCP tools, allowing LLMs and other MCP clients to:

Primary Tools:

• get_text - Retrieve Jewish texts by reference (e.g., "Genesis 1:1")
• text_search - Search across the entire Jewish library
• getcurrentcalendar - Get situational Jewish calendar information. Accepts diaspora (true for Diaspora, false for Israel) for weeks when Torah reading schedules diverge.
• englishsemanticsearch - Semantic similarity search on English text embeddings (Currently only works from official Sefaria MCP)

Core Tools:

• getlinksbetween_texts - Find cross-references and connections between texts
• searchinbook - Search within a specific book or text work
• searchindictionaries - Search Jewish reference dictionaries

Support Tools:

• getenglishtranslations - Retrieve all available English translations for a text
• gettopicdetails - Retrieve detailed information about topics in Jewish thought
• clarifynameargument - Autocomplete and validate text names, book titles, and topics
• clarifysearchpath_filter - Convert book names to proper search filter paths

Structure Tools:

• gettextorcategoryshape - Explore the hierarchical structure of texts and categories
• gettextcatalogue_info - Get bibliographic and structural information (index) for a work

Manuscript Tools:

• getavailablemanuscripts - Access historical manuscript metadata and image URLs
• getmanuscriptimage - Download and process specific manuscript images

All endpoints are optimized for LLM consumption (compact, relevant, and structured responses).

What is MCP?

MCP (Model Context Protocol) is an open protocol for connecting Large Language Models (LLMs) to external tools, APIs, and knowledge sources. It enables LLMs to retrieve, reference, and interact with structured data and external services in a standardized way. Learn more in the MCP documentation.

How to Run

Prerequisites

• Python 3.10+
• Docker (optional, for containerized deployment)

Local Development

1. Install dependencies:

``bash pip install -e . ``

1. Run the server:

``bash python -m sefaria_mcp.main ` The server will be available at http://127.0.0.1:8088/sse by default. Set SEFARIAMCPPORT to override the SSE/API port (e.g., SEFARIAMCPPORT=8089 python -m sefariamcp.main). Prometheus metrics bind separately on SEFARIAMCPMETRICSPORT (default 9090`).

Docker

1. Build the image:

``bash docker build -t sefaria-mcp . ``

1. Run the container:

``bash docker run -d --name sefaria-mcp \ -e SEFARIA_MCP_PORT=8089 \ -e SEFARIA_MCP_METRICS_PORT=9090 \ -p 8089:8089 \ -p 9090:9090 \ sefaria-mcp ` The server will be available at http://localhost:8089/sse and metrics at http://localhost:9090/` (adjust the port mappings as needed).

Usage

• Connect your MCP-compatible client to the /sse endpoint.
• All tool endpoints are available via the MCP protocol.

Monitoring

• Prometheus metrics are exposed via the standalone HTTP server started on SEFARIA_MCP_METRICS_PORT (defaults to 9090).
• Scrape http://localhost:9090/ (or your configured host/port). Metrics include:
• mcp_tool_calls_total{tool_name,status} – call counts per tool and status.
• mcp_tool_duration_seconds{tool_name} – histogram of per-call durations.
• mcp_tool_payload_bytes{tool_name} – histogram of response payload sizes.
• mcp_errors_total{tool_name,error_type} – per-tool error counts.
• mcp_active_connections – current SSE connection gauge.
• Standard FastAPI instrumentation (request rate, latency, status codes, in-progress requests, etc.) from prometheus_fastapi_instrumentator.

Commit Hygiene

This repo uses semantic commits with the fix, feat, and chore keywords.

Acknowledgments

Special thanks to @Sivan22 for pioneering the first Sefaria MCP server (mcp-sefaria-server), which inspired this project and the broader effort to make Jewish texts accessible to LLMs and modern AI tools.

License

MIT

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Sefaria
• Source: Sefaria/sefaria-mcp
• License: MIT

Install and usage instructions live in the source repository linked above.