Openarx MCP server
{
"mcpServers": {
"openarx": {
"type": "http",
"url": "https://mcp.openarx.ai/researcher/mcp"
}
}
}{
"mcpServers": {
"openarx": {
"type": "http",
"url": "https://mcp.openarx.ai/researcher/mcp"
}
}
}{
"mcpServers": {
"openarx": {
"type": "http",
"url": "https://mcp.openarx.ai/researcher/mcp"
}
}
}{
"servers": {
"openarx": {
"type": "http",
"url": "https://mcp.openarx.ai/researcher/mcp"
}
}
}{
"mcpServers": {
"openarx": {
"type": "http",
"url": "https://mcp.openarx.ai/researcher/mcp"
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsonversion 0.3.2
What it does
OpenArx exposes a knowledge graph of scientific and engineering claims connected by typed relations, designed for AI agents to search, reason over, and contribute to the knowledge record. It is for developers building AI agents that need to read and write structured scientific knowledge rather than unstructured documents.
- Search across scientific papers using hybrid semantic and keyword search to find claims and evidence relevant to a research question.
- Traverse a graph of typed relations (support, extend, qualify, refute) between claims to understand how knowledge connects.
- Publish new claims and relations back into the graph through a methodology engine that reviews contributions stage-by-stage before they reach the knowledge base.
Tools this server exposes
| Tool | Does |
|---|---|
| search | Hybrid semantic and keyword search across scientific papers. |
| search_keyword | Pure keyword (BM25) search for exact-term lookups. |
| search_semantic | Pure vector search for paraphrased queries and concept exploration. |
| find_related | Find related papers by similarity, entity, or concept. |
| paginate | Continue a previous search from its cached candidate pool. |
| find_evidence | Check a claim against the corpus and group supporting or contradicting evidence. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.