BrowseAI Dev MCP Server

BrowseAI Dev

Research infrastructure for AI agents — real-time web search, evidence extraction, and structured citations. Every claim is backed by a URL. Every answer has a confidence score.

Agent → BrowseAI Dev → Internet → Verified answers + sources

Website · Playground · API Docs · Alternatives · Discord

Package names: npm: `browseai-dev` · PyPI: `browseaidev` · LangChain: `langchain-browseaidev` — Previously browse-ai and browseai. Old names still work and redirect automatically.

How It Works

search → fetch pages → neural rerank → extract claims → verify → cited answer (streamed)

Every answer goes through a multi-step verification pipeline. No hallucination. Every claim is backed by a real source.

Verification & Confidence Scoring

Confidence scores are evidence-based — not LLM self-assessed. After the LLM extracts claims and sources, a post-extraction verification engine checks every claim against the actual source page text:

1. Atomic claim decomposition — Compound claims are auto-split into individual verifiable facts. "Tesla had $96B revenue and 1.8M deliveries" becomes two atomic claims, each verified independently.
2. Hybrid retrieval (BM25 + dense embeddings) — For each claim, BM25 finds lexical matches and OpenAI text-embedding-3-small (via OpenRouter) finds semantic matches from source text. Rankings are fused using Reciprocal Rank Fusion (RRF) — a rank-based method that avoids score normalization issues. This catches paraphrased evidence that BM25 alone misses (e.g., "prevents fabricated answers" matching "reduces hallucinations"). Premium tier only, with graceful BM25 fallback.
3. NLI evidence reranking — Top-3 RRF-fused candidates per claim are reranked by a DeBERTa-v3 NLI model for semantic entailment. Final hybrid score: 30% BM25 + 70% NLI, with contradiction penalties and paraphrase boosts.
4. Multi-provider search — Parallel search across multiple providers for broader source diversity. More independent sources = stronger cross-reference = higher confidence.
5. Domain authority scoring — 10,000+ domains across 5 tiers (institutional .gov/.edu → major news → tech journalism → community → low-quality), stored in Supabase with Majestic Million bulk import. Self-improving via Bayesian cold-start smoothing.
6. Source quote verification — LLM-extracted quotes verified against actual page text using hybrid matching (exact substring → BM25 fallback).
7. Cross-source consensus — Each claim verified against all available page texts. Claims supported by 3+ independent domains get "strong consensus". Single-source claims flagged as "weak".
8. Contradiction detection — Claim pairs analyzed for semantic conflicts using topic overlap + NLI contradiction classification. Detected contradictions surfaced in the response and penalize confidence.
9. Multi-pass consistency — In thorough mode, claims are cross-checked across independent extraction passes. Claims confirmed by both passes get boosted; inconsistent claims are penalized (SelfCheckGPT-inspired).
10. Auto-calibrated confidence — 7-factor confidence formula auto-adjusts from user feedback using isotonic calibration curves. Predicted confidence aligns with actual accuracy over time. Factors: verification rate (25%), domain authority (20%), source count (15%), consensus (15%), domain diversity (10%), claim grounding (10%), citation depth (5%).
11. Per-claim evidence retrieval — Weak claims get targeted search queries generated by LLM, then searched individually across all providers. Each claim gets its own evidence pool instead of sharing the same corpus (SAFE-inspired, from Google DeepMind's fact-checking research).
12. Counter-query verification — Verified claims are stress-tested with adversarial "what would disprove this?" search queries. If counter-evidence is found, claim confidence is penalized (SANCTUARY-inspired).
13. Iterative confidence-gated retrieval — Thorough mode uses a FIRE-inspired loop: verify → if weak claims remain → generate