Research Connector Hub
MCP server that turns operator-authorized local or synthetic source records into auditable research briefs, with every finding linked to exact excerpts and hashes. Exposes a single build_research_brief tool over deterministic, no-network stdio.
README
Research Connector Hub
Research Connector Hub turns operator-authorized local or synthetic source records into a brief that another person can audit. Every finding links to an exact excerpt, character offset, source hash, and excerpt hash.
It is not a web scraper, an LLM summarizer, or an Apify wrapper. The default workflow performs no network requests, needs no token, emits no telemetry, and has no runtime package dependencies.
Five-minute demo
Requirements: Node.js 22 or newer.
npm test
npm run demo
The demo writes four files to demo-output/:
brief.md: human-readable findings and citations.brief.json: structured brief and source catalog.evidence.json: exact excerpts, offsets, and hashes.run.json: deterministic run ID, config hash, connector receipts, counts, and warnings.
Use the included local-data example:
node ./bin/research-hub.js run \
--config ./examples/market-signals/research.json \
--output ./research-output
Research specification
{
"title": "Synthetic market signals",
"question": "Which qualities appear in the supplied notes?",
"queries": ["workflow automation", "evidence traceability"],
"maxEvidencePerQuery": 2,
"connectors": [
{ "type": "local-json", "path": "sources.json" }
]
}
Local paths are relative to the config file. Absolute paths, directory traversal, symlinked inputs, directories owned by another non-root user, invalid UTF-8 or malformed Unicode strings, configs above 1 MB, source files above 5 MB, aggregate local input above 10 MB, more than 1,000 raw local records, more than 10 connectors, more than 200 documents, aggregate document content above 10 MB, token-dense queries or sources, and documents above 500 KB fail closed. URL query parameters are rejected so signed or credential-bearing locators cannot enter artifacts. Local file connectors and output generation require POSIX ownership/mode checks. Output directories and artifacts are private to the current user (0700/0600).
Source records use this shape:
{
"title": "Authorized source",
"url": "https://example.test/source",
"content": "Exact source text.",
"publishedAt": "2026-08-01T00:00:00.000Z",
"sourcePolicy": {
"license": "License or internal-use basis",
"termsUrl": "https://example.test/terms",
"authorizedBy": "source owner or operator"
}
}
If both sourcePolicy.license and sourcePolicy.termsUrl are missing, the run records a warning. Policy metadata is never treated as proof of authorization.
Connectors
mock: deterministic synthetic records for evaluation and demos.local-json: operator-supplied JSON records, filtered locally by query terms.
Future connectors implement one bounded collect({ query, limits, signal }) contract and return both documents and a provider receipt. See ARCHITECTURE.md and COMMERCIAL_BOUNDARIES.md.
MCP stdio adapter
node ./bin/research-hub.js mcp
The adapter exposes one tool, build_research_brief, over newline-delimited JSON-RPC stdio. It accepts inline documents only and calls the same deterministic pipeline. Input frames are capped while streaming and decoded as strict UTF-8, request IDs are bounded scalars, notifications receive no response, output is capped and backpressured, and Markdown is returned once as text rather than duplicated in structured content. It implements a tested subset of the MCP 2025-11-25 surface (initialize, tools/list, tools/call); it does not claim full MCP conformance.
MCP is an adapter, not the product core. The specification describes tools and structured output, while the evidence and research semantics remain application responsibilities. See the MCP 2025-11-25 specification.
Apify boundary
The project was designed after auditing Apify MCP Server, but contains no copied upstream source, credentials, telemetry keys, images, cache, or Git history. Apify can be added later as an optional connector only with explicit Actor selection, spend limits, source-policy acknowledgement, terminal run evidence, and provider cost receipts.
Apify pricing varies by Actor and can include compute, storage, proxies, transfer, per-event charges, or rental fees. Review the current Apify pricing, Store Actor pricing models, and applicable terms before enabling an external connector.
Verification
npm test
npm run lint
npm run build
npm run check:package
npm run build performs an npm package dry run because the product ships native Node.js ESM and has no compile step.
Status
This is an MVP for product validation. It produces candidate findings from exact source sentences; it does not verify truth, resolve contradictory evidence, or replace human review.
License: MIT. See NOTICE and THIRD_PARTY_NOTICES.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.