# BrowseAI-HQ/BrowserAI-Dev [Health: Active]

**Category:** 🔬 Research  
**Repository:** https://github.com/BrowseAI-HQ/BrowserAI-Dev  
**GitHub Stars:** 20  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/browseai-hq-browserai-dev

## Description
Evidence-backed web research for AI agents. Real-time search with cited claims, confidence scores, and compare mode (raw LLM vs evidence-backed). MCP server, REST API, and Python SDK.

## 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 evidence-backed answer
- **clarity** — Anti-hallucination answer engine — 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": {
  "browserai-dev": {
    "command": "npx",
    "args": ["-y","lastsearch"],
    "env": {
      "LASTSEARCH_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `LASTSEARCH_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What BrowseAI-HQ/BrowserAI-Dev MCP server does

BrowseAI-HQ/BrowserAI-Dev MCP server exposes web research functions to MCP-compatible agents. The available tools cover the full path from discovery to evidence-backed output: `search` finds relevant web information, `open` retrieves a page as cleaned text, and `extract` identifies structured claims from page content.

The `answer` tool combines those operations into a cited response. It accepts a `depth` value of `fast`, `thorough`, or `deep`, allowing callers to choose between a quick lookup and more extensive verification. `compare` places a raw LLM response beside an evidence-backed response, while `clarity` supports prompt-only, LLM-generated, and web-fused verified modes.

## How it works

BrowseAI-HQ/BrowserAI-Dev MCP server follows a research pipeline that searches for sources, fetches pages, extracts claims, and checks those claims against source text. Verification can use keyword matching, semantic retrieval, source authority, cross-source agreement, quote checking, contradiction detection, and other evidence signals described by the project. The returned data includes claim and source fields such as verification status, verification score, consensus information, and authority information.

The depth setting changes the amount of work performed. `fast` uses one search, extraction, and verification pass. `thorough` can repeat the process for weak claims, retrieve evidence for individual claims, and perform counter-query and consistency checks. `deep` adds iterative research steps and gap analysis, but requires a LastSearch key and sign-in; without that key or when quota is exhausted, it falls back to thorough mode.

The server also supports persistent research sessions. `session_create` starts a session, `session_ask` researches within it, and `session_recall` queries stored knowledge without starting a new web search. Sessions can be exported, shared publicly, or forked for continued work.

## Setup and configuration

The project identifies its packages as `lastsearch` on npm and PyPI. It also documents a REST API and Python SDK in addition to the MCP server. The supplied material does not provide a complete MCP client configuration or a confirmed command for launching the MCP process, so installation commands should be taken from the project’s current documentation rather than inferred here.

Session features require a LastSearch API key. For MCP usage, the documented environment variable is `LASTSEARCH_API_KEY`; REST clients use a bearer token, and the Python SDK accepts an `api_key` argument. A free key is available through the LastSearch dashboard. Deep mode and some semantic verification capabilities are described as premium features, with keyword-only verification available as a fallback when premium keys are not configured.

## Tools and capabilities

- Search the web with `search`.
- Fetch and clean page content with `open`.
- Extract structured claims with `extract`.
- Produce cited, verified answers with `answer`.
- Compare raw and evidence-backed responses with `compare`.
- Select prompt, answer, or verified behavior with `clarity`.
- Create, query, recall, share, export, and fork research sessions.
- Submit result feedback through `feedback`.

## Limitations and notes

BrowseAI-HQ/BrowserAI-Dev MCP server depends on external web search, page-fetching, and language-model services, so results depend on source availability and provider responses. The project automatically retries transient failures such as rate limits and server errors, but authentication failures do not receive the same retry treatment.

Evidence-backed output is not identical to certainty. The system reports confidence and exposes verification details so an agent can apply its own trust policy. Single-source claims may be marked weak, while contradictions can be surfaced in the response. Deep mode has higher resource and quota requirements, including a documented cost of three times the quota per query and a limit of 100 deep queries per day.

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