BookMCP v0.1.0 — Book knowledge for system design

BookMCP v0.1.0

Give coding agents the book knowledge to design and build better systems.

BookMCP brings your system design and engineering books into your agent's workflow. Codex, Claude Code, or another MCP client can retrieve relevant principles, compare architecture trade-offs, and use cited evidence to guide implementation. Review and save a source-linked design lesson through the CLI, then retrieve it in a later session.

For example: ask your agent to search your books for retries and idempotency, compare approaches for a background job service, and propose interfaces and failure-case tests. BookMCP supplies the evidence; the agent still needs to reason about the workload and validate the code. It does not train the model.

Watch the 45-second CLI/MCP demo, recorded from real output with an original programming guide. The source and transcript are included.

Download

No Rust toolchain, Python runtime, API key, or separate database is needed for the native binaries.

Platform	Archive
macOS Apple Silicon	ARM64
macOS Intel	x64
Linux x86-64	x64
Linux ARM64	ARM64
Windows x86-64	x64

Download the matching .sha256 file to verify your archive. Extract it and follow the included installation guide. The archives include an original sample PDF, documentation, and dependency licenses. macOS binaries are not Apple-notarized; Linux binaries are built on Ubuntu 22.04 with glibc.

For clients that support MCPB installation, the portable MCP bundle includes all five executables. Select the library directory reported by bookmcp doctor during setup. CLI ingestion remains the way to add books.

Try it

From the extracted folder, with bookmcp on your PATH:

bookmcp ingest tests/fixtures/tiny.pdf --book-id tiny-test --title "Tiny Test Book"
bookmcp search "citations" --book-id tiny-test
bookmcp doctor

Connect your agent:

codex mcp add bookmcp -- bookmcp serve
# Or:
claude mcp add --transport stdio --scope user bookmcp -- bookmcp serve

In a new agent session: “Use BookMCP to show my library. Search tiny-test for citations, read the matching chunk, and explain its advice with a page citation.”

Included in this release

• Multi-book PDF ingestion, deterministic chunks, managed originals, and 1-based PDF page citations.
• Compact library discovery, keyword search, bounded context, read-only MCP tools/resources, and prompts.
• Explicit CLI lesson saves with source references and stale-source detection.
• Configuration helpers for Codex and Claude, integrity diagnostics, and index repair.
• Validation of IDs and retrieval limits, encrypted/scanned PDF errors, and cross-platform CI.

OCR, semantic/vector search, and automatic model memory are not implemented. MCP can draft and read lessons; saves happen explicitly through the CLI. BookMCP makes no hosted AI calls. Retrieved text enters the connected agent's context, so use a local model/client for an entirely offline assistant.

Please share first-run feedback: your platform, the first confusing step, and whether you reached a cited result. Leave private PDF text out of reports.