# BrowseAI Dev [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/BrowseAI-HQ/BrowseAI-Dev  
**GitHub Stars:** 20  
**npm Downloads (last month):** 389  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/browseai-dev

## Description
Evidence-backed web research for AI agents with citations and confidence scores.

## Tools
Capabilities this server exposes over MCP:

- **search** — Search the web for information on any topic
- **open** — Fetch and parse a web page into clean text
- **extract** — Extract structured claims from a page
- **answer** — Full pipeline: search + extract + cite. `depth`: `"fast"`, `"thorough"`, or `"deep"
- **compare** — Compare raw LLM vs source-checked answer
- **clarity** — Answer mode that aims to reduce hallucinations — three modes: prompt (prompts only), answer (LLM), verified (LLM + web fusion)
- **session_create** — Create a research session (persistent memory)
- **session_ask** — Research within a session (recalls prior knowledge)
- **session_recall** — Query session knowledge without new web search
- **session_share** — Share a session publicly (returns share URL)
- **session_knowledge** — Export all claims from a session
- **session_fork** — Fork a shared session to continue the research
- **feedback** — Submit feedback on a result to improve accuracy

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "browseai-dev": {
    "command": "npx",
    "args": ["-y","verify-anything"]
  }
}
```

## Documentation

## What the BrowseAI Dev MCP server does

BrowseAI Dev MCP server exposes web research and verification capabilities to an MCP-compatible AI agent. The repository documentation identifies the underlying research system as LastSearch, which searches the internet, retrieves pages, extracts claims, and produces answers with source URLs and confidence information.

The server is suited to questions where an agent needs to show its evidence. It can return structured claims, source details, consensus signals, contradiction findings, and citation health information. Agents can use those fields to make their own decisions about whether an answer is sufficiently supported.

## How it works

The core research flow starts with web search, fetches and parses relevant pages, extracts claims, and verifies those claims against the source text. The `answer` operation combines these stages and can run at three depths:

- `fast` performs a single search, extraction, and verification pass.
- `thorough` can run additional evidence retrieval, counter-query checks, and consistency checks for weak claims.
- `deep` performs iterative research passes that identify gaps, generate follow-up queries, and re-check the combined result. It requires a LastSearch key and sign-in; without the key or after quota exhaustion, it falls back to thorough mode.

Verification can include keyword and semantic matching, source authority, cross-source agreement, contradiction detection, quote checking, and citation URL health. The documentation says premium keys enable semantic retrieval and reranking, while keyword-only verification remains available without those keys.

## Tools and capabilities

BrowseAI Dev MCP server provides separate tools for each stage as well as higher-level research workflows:

- `search` finds information on a topic.
- `open` retrieves a web page and converts it to clean text.
- `extract` identifies structured claims from a page.
- `answer` returns a researched answer with extracted evidence and citations.
- `compare` contrasts a raw language-model answer with an evidence-backed answer.
- `clarity` supports prompt-only, language-model, and verified web-fusion modes.
- Session tools create, query, recall, share, export, and fork persistent research sessions.
- `feedback` submits feedback about a result for accuracy improvement.

Session research retains verified knowledge for later questions, so follow-up queries can use prior findings without performing a new web search. Shared sessions return a public share URL, while exported session knowledge provides the accumulated claims.

## Limitations and notes

BrowseAI Dev MCP server depends on external web pages and search, language-model, and page-fetching services. Those calls may encounter transient rate limits or server errors; the documented system retries such failures with exponential backoff, while authentication errors fail immediately.

Deep research uses more quota than the other modes: the documentation states that each deep query costs three times the quota and that deep mode supports 100 queries per day. Deep research targets a confidence threshold of 0.85 and can run up to four total research steps. Session features require a LastSearch API key for identity and ownership.

Confidence is evidence-based rather than a language model's self-assessment, but it remains a signal for agent decision-making, not a substitute for reviewing the cited material. A claim supported by only one source may be marked weak, while conflicting sources can reduce confidence. Citation health distinguishes live, stale, and dead URLs when the relevant checks are available.

_Full upstream README: https://allmcps.com/mcp/browseai-dev/readme_

