Your AI already knows find, grep, and cat. Give it a virtual filesystem over your docs — plus semantic search when it needs meaning, not just matches. Self-hosted. Open source. Zero lock-in.
server: name: my-project-docs sources: - name: docs type: markdown repo: github.com/you/project path: docs/ tools: - name: search-docs type: search source: docs - name: explore-docs type: bash sources: [docs] bash: session_state: true grep_strategy: hybrid
Agents choose the right tool for each sub-task — no other MCP server does both
pgvector embeddings with cosine similarity. Ask conceptual questions: "how do I configure auth?"
find, grep, cat, ls, head — agents explore docs like a developer would. Sandboxed via just-bash.
cd persists across commands. Agents navigate naturally: cd docs/guides && ls && cat auth.mdx
3-pass search: semantic embeddings + ILIKE text + dedup. Falls back gracefully when infra is down.
`related /path` finds similar files across sources. Grep misses suggest companion search tools.
Bash-only tools work with no database, no API keys. Add RAG when you're ready — it's just config.
Mintlify replaced RAG with a filesystem. We compose both.
| Capability | Pathfinder | ChromaFS |
|---|---|---|
| Semantic search | ✓ pgvector RAG | ✓ Via Chroma |
| File exploration | ✓ find/grep/cat/ls | ✓ find/grep/cat/ls |
| Both paradigms composable | ✓ | ✗ FS-only |
| Session state (persistent CWD) | ✓ | ✗ |
| Vector-backed grep | ✓ Configurable | ✓ Chroma-backed |
| Writable workspace | ✓ | ✗ EROFS |
| Multi-source | ✓ Config-driven | ✗ Single collection |
| Feedback collection | ✓ Collect tools | ✗ |
| Zero-infra mode | ✓ Bash-only | ✗ Needs Chroma+Redis |
| Open source | ✓ MIT | ✗ Proprietary |
| Auto-refresh on webhook | ✓ | ~ Manual reingestion |
Your docs are already structured for it. Our migration guide takes 15 minutes.
Migration Guide →