Tripitaka MCP

What dhamma-seeker/tripitaka-mcp MCP server does

The dhamma-seeker/tripitaka-mcp MCP server exposes the Pāli Tipiṭaka to MCP-compatible AI clients. Its indexed collection covers approximately 444,000 segments across the Sutta, Vinaya, and Abhidhamma baskets. Sutta material includes Pāli and available English translations; the Abhidhamma data is Pāli-only because the upstream source does not provide English for those books.

The server supports research tasks such as finding passages by topic, retrieving a complete text by identifier, comparing translations, and checking Pāli vocabulary. Responses include links to a bilingual reader with segment anchors, allowing users to inspect the cited passage. The project is offered free for non-commercial use as Dhamma Dāna. Its code is MIT-licensed, while the repository separately documents data terms.

How it works

The hosted service provides Streamable HTTP at https://mcp.tripitaka-mcp.com/mcp and legacy SSE at https://mcp.tripitaka-mcp.com/sse. Hosted search uses PostgreSQL with pgvector. Keyword search uses trigram matching, semantic search uses vector similarity, and hybrid search combines both result types with Reciprocal Rank Fusion.

The local edition uses a bundled SQLite database. It can run without network access after its initial database download, but it does not include the hosted semantic or hybrid search tools. Local keyword search uses SQLite FTS5, so its matching and ranking may differ from the hosted PostgreSQL implementation. The local database is a snapshot and must be refreshed manually.

Setup and configuration

For offline use, install the package with Python 3.10 or newer:

pipx install tripitaka-mcp
tripitaka-mcp init
tripitaka-mcp serve

The init command downloads the approximately 120 MB SQLite database. MCP clients can launch the resulting tripitaka-mcp serve command over stdio. An HTTP mode is also available locally with tripitaka-mcp serve --http; the default address is 127.0.0.1:8765/mcp, and the port can be changed with --port or the documented MCP_PORT setting. The hosted option requires no local database or Docker setup.

Tools and capabilities

The dhamma-seeker/tripitaka-mcp MCP server provides capabilities for:

• Searching by keyword, semantic similarity, or a hybrid ranking method.
• Fetching texts by IDs such as mn1, pli-tv-bu-vb-pj1, and patthana1.1.
• Producing formatted academic citations for retrieved passages.
• Comparing translations at aligned segment boundaries.
• Looking up Pāli words in a dictionary containing more than 20,000 entries from P. A. Payutto, PTS, and DPPN sources.
• Reducing inflected forms to possible roots when a direct dictionary lookup fails.
• Reading structure and text data through MCP resources such as tripitaka://structure, tripitaka://sutta/{id}, and tripitaka://word/{w}.

The repository also includes curated reference pages, a Claude research workflow file, and a Postman collection, but these are supporting project assets rather than separate canon-search methods.

Limitations and notes

The public service is rate-limited to 10 requests per 10 seconds and 60 requests per minute per IP. Queries sent to the hosted service leave the local machine; the project provides a privacy policy for that deployment. Local mode avoids network requests after initialization and has no hosted rate limit.

Local mode exposes fewer tools than the hosted deployment: semantic and hybrid search are unavailable, and the trigram index is replaced by SQLite FTS5. A local HTTP listener binds to localhost by default. Anyone exposing it more broadly should consider that the server does not authenticate callers. Commercial use is not permitted under the project’s stated Dhamma Dāna terms.