jev-research-mcp
Self-hosted MCP research server that filters search results and page blocks with Jev, returning compact evidence with provenance.
Overview
JEV Research MCP is a self-hosted, local-first MCP server for coding-agent web research. Instead of fetching 20 full search results into an expensive model's context, it filters twice with Jev: once at the search-result level before fetching, and once at the content-block level after parsing, and returns compact evidence blocks with provenance. It selects evidence and never rewrites, summarises or generates content.
The single research tool runs a Brave search, sends one state with relevant, useful and authoritative questions per result in batches, and keeps results that are both relevant and useful. Selected pages are fetched with bounded concurrency, SSRF checks and size, timeout and content-type gates, then parsed deterministically into headings, paragraphs, code, lists, quotes and tables. A second Jev pass asks whether each block is relevant and actionable. Assembly restores heading context, deduplicates and applies a token budget. The principle is that Jev decides and code disposes.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this search result relevant / useful / authoritative for the task? | yes · no |
| Yes / no | Is this content block relevant and actionable? | yes · no |
How it works
- Search with Brave and normalise results
- Jev filters results before fetching
- Fetch and parse selected pages into blocks
- Jev filters blocks for relevance and actionability
- Assemble deduplicated evidence within a token budget
Strengths
- Saves agent context and avoids fetching discarded pages
- Evidence returned verbatim with provenance
- Hardened fetching with SSRF checks
Caveats
- Requires Brave Search and TypeSafe keys
- Conservative filtering may drop useful content
At a glance
- Type
- MCP server
- Status
- beta
- Runtime
- MCP server
- Has code
- yes
Use cases and integrations
How curated lists describe it
JEV Evidence MCP is a self-hosted, local-first Model Context Protocol (MCP) (https://modelcontextprotocol.io) server for coding-agent web research. Instead of dumping 20 full search results into an expensive downstream LLM context window, it selects evidence semantically with TypeSafe JEV.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which search results to fetch and which page blocks are relevant and actionable.
Does it summarise pages?
No. It returns selected evidence blocks verbatim.
Which search provider does it use?
Brave Search API.
Related projects
MCP server that scores every search query, result and fetched page an agent sees for relevance, credibility and steering risk.
MCP serverbetaTypeScriptMCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
MCP serverbetaTypeScriptnpm CLI that screens large agent tool outputs with Jev, returning compact typed decisions and IDs while originals stay recoverable.
CLIbetaTypeScriptExperimental Rust SEO and GEO CLI and MCP server combining page checks, DuckDuckGo queries and optional Jev scoring.
MCP serverprototypeRustMCP tool that asks Jev which files, web pages or snippets are relevant to a query, so the agent reads selectively.
MCP serverbetaTypeScriptMCP server with record and recall tools: facts are logged during a session and Jev keeps only those relevant to the next task.
MCP serverbetaTypeScript