Falsafa MCP

Stdio MCP server for the Falsafa corpus. Ten librarian-flavored tools so any LLM client (Claude Desktop, Claude Code, Cursor, Codex, or any MCP-aware host) can navigate 37 translated philosophical and classical works through paragraph-stable citations. No API key, no setup beyond npx.

npx -y @falsafa/mcp

Install from the npm registry, not from a git URL — git installs trigger the package's prepack hook, which depends on the source tree's corpus/ directory and bun. Registry installs ship the corpus pre-bundled in the tarball.

First run downloads ~48 MB (the corpus ships inside the tarball). If your MCP client's startup timeout is short — Claude Code in particular — run npx -y @falsafa/mcp once in a terminal first. npm caches the package, and your client's spawn resolves instantly thereafter.

Install in your daily LLM

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "falsafa": { "command": "npx", "args": ["-y", "@falsafa/mcp"] }
  }
}

Restart Claude Desktop. The Falsafa tools show up in the tool palette. Ask "what works does Cynewulf have?" and the model calls list_works({ author: "cynewulf" }).

Claude Code

claude mcp add falsafa npx -y @falsafa/mcp

Cursor

Settings → MCP → Add new global MCP server, paste:

{
  "mcpServers": {
    "falsafa": { "command": "npx", "args": ["-y", "@falsafa/mcp"] }
  }
}

Or edit ~/.cursor/mcp.json directly with the same shape.

Codex CLI

codex mcp add falsafa -- npx -y @falsafa/mcp

Persists in ~/.codex/config.toml. The -- separator is required.

Any other stdio MCP client

The universal config shape is:

{ "command": "npx", "args": ["-y", "@falsafa/mcp"] }

Drop that wherever your client expects an MCP server entry.

Tools

Ten tools. Eight for catalog navigation, two for the rule-based wiki layer.

• list_works — list works in the corpus with optional author / era / genre / language filters
• list_chapters — list chapters of a specific work
• get_metadata — full metadata + variant counts for a work
• read_chapter — full chapter text. Body is annotated with [p-XXXXXX] paragraph-id markers; use those for paragraph citations.
• get_passage — read specific paragraphs by id list or 0-indexed range. Each result has a citation_url ready to drop into a markdown footnote.
• search_corpus — search English bodies. Distinctive 2-3 word phrases work best.
• find_related — TF-IDF-based related chapters, with a structural fallback.
• compare_works — side-by-side pointer chapters for two works on a topic.
• read_wiki — rule-based wiki card (~280 tokens) for a work or chapter. Use BEFORE read_chapter to scan what's worth a deep read. Cards are deterministic, generated from the corpus by classical statistical algorithms — zero LLM tokens in any output. Each card includes verbatim openings, closings, and key passages with [p-XXXXXX] cite handles.
• read_wiki_full — heavier wiki sheet (~1,500 tokens) with the deeper statistical detail layered on top of the card. Opt-in for deep analysis; most queries should use read_wiki first and only escalate when needed.

What's in the corpus

37 works spanning Old English Christian poetry (Cynewulf), Urdu ghazal masters (Ghalib, Iqbal, Zauq), French Enlightenment political theory (Comte, Dunoyer), German philosophical writing (Fichte), Sanskrit smṛti traditions, and Old Javanese / Kawi tattva texts. Each work ships with the original-language source, a Latin-script transliteration where it makes sense, and an English translation. Every paragraph has a stable content-derived ID (p-xxxxxx) so citations survive reformatting.

Translations and transliterations are AI-assisted. AI can make mistakes — when accuracy matters, verify against the original-language source linked on each chapter page. Translations are produced by Thothica's pipeline across Claude / GPT / Gemini. Underlying source archives:

• Old English (Cynewulf, OE Elegies) — sacred-texts.com
• Sanskrit smṛti corpus — GRETIL, Göttingen Register of Electronic Texts in Indian Languages
• Allama Iqbal (Bāng-i-Darā) — allamaiqbal.com, Iqbal Academy Pakistan
• Mirza Ghalib + Sheikh Ibrahim Zauq — printed editions

Full source acknowledgments at falsafa.ai/about/#sources.

Links

• falsafa.ai — reading site, eval explorer, thesis on why this design
• falsafa.ai/thesis/#methodology — how eval scoring works (deterministic citation check; no LLM judge)
• GitHub — adoistic/falsafa for source

License

MIT.