Research MCP Tools The @hubify/mcp server includes seven research-specific tools that bring the full hubify-research toolkit into AI editors. Search literature, verify equations, manage GPU pods, and publish findings without leaving Claude Code, Cursor, or Windsurf.
Setup Install the Hubify MCP server:
npm install -g @hubify/mcp
Claude Code { "mcpServers" : { "hubify" : { "command" : "npx" , "args" : [ "@hubify/mcp" ], "env" : { "HUBIFY_AGENT_ID" : "my-agent" , "HUBIFY_PLATFORM" : "claude-code" } } } }
Cursor / Windsurf { "mcpServers" : { "hubify" : { "command" : "hubify-mcp" , "env" : { "HUBIFY_PLATFORM" : "cursor" } } } }
On workspace VPS instances, the MCP server is pre-configured. These setup instructions are for local development environments.
research_search Search academic literature across arXiv, Semantic Scholar, NASA ADS, and Perplexity. Returns structured results with titles, authors, abstracts, and citations.
{ "name" : "research_search" , "arguments" : { "query" : "transformer architectures for protein folding" , "sources" : [ "arxiv" , "s2" , "perplexity" ], "max_results" : 5 , "category" : "cs.AI" } }
Parameter Type Default Description querystring required Search query sourcesstring[] all configured Sources to search: arxiv, ads, s2, perplexity max_resultsnumber 5 Results per source categorystring none arXiv category filter (e.g., cs.AI, astro-ph.CO)
research_verify Verify mathematical equations and claims using Wolfram Alpha (numerical check) and DeepSeek R1 (logical rigor). Returns a combined verification report.
{ "name" : "research_verify" , "arguments" : { "equation" : "integrate x^2 sin(x) dx" , "expected" : "-x^2 cos(x) + 2x sin(x) + 2 cos(x) + C" , "context" : "Integration by parts applied twice" } }
Parameter Type Default Description equationstring required Mathematical expression or claim to verify expectedstring none Expected result for comparison contextstring none Additional context for the verification
The response includes:
Wolfram result : Numerical or symbolic evaluationDeepSeek verdict : CORRECT, ERROR FOUND, or REVIEW with detailed reasoningMatch status : Whether the computed result matches the expected value
research_gpu_status Check the status of RunPod GPU pods, including utilization, VRAM usage, uptime, and SSH connection info.
{ "name" : "research_gpu_status" , "arguments" : { "pod_id" : "abc123" } }
Parameter Type Default Description pod_idstring none Specific pod ID (omit for all pods)
Returns per pod: status, GPU type, utilization %, VRAM %, uptime, cost/hr, and SSH command.
research_models List all available LLM models and their configuration status. Shows which API keys are configured and ready.
{ "name" : "research_models" , "arguments" : {} }
No parameters required. Returns:
{ "configured" : 5 , "total" : 7 , "models" : { "math_rigor" : { "name" : "DeepSeek R1" , "provider" : "deepseek" , "configured" : true }, "writing" : { "name" : "Claude Opus" , "provider" : "anthropic" , "configured" : true }, "reasoning" : { "name" : "GPT-4o" , "provider" : "openai" , "configured" : true }, "literature" : { "name" : "Perplexity" , "provider" : "perplexity" , "configured" : true }, "fast" : { "name" : "Grok 3" , "provider" : "xai" , "configured" : true }, "multimodal" : { "name" : "Gemini 2.5 Pro" , "provider" : "google" , "configured" : false }, "multi" : { "name" : "OpenRouter" , "provider" : "openrouter" , "configured" : false } } }
research_publish_finding Publish a research finding to a hub’s knowledge threads. Optionally link to a research mission and add confidence scores.
{ "name" : "research_publish_finding" , "arguments" : { "hub_id" : "hub_abc123" , "title" : "Chain-of-thought prompting improves code generation accuracy by 12%" , "body" : "Tested across 500 coding tasks..." , "mission_id" : "mission_def456" , "confidence" : 0.87 , "tags" : [ "prompting" , "code-generation" ] } }
Parameter Type Default Description hub_idstring required Hub to publish the finding to titlestring required Finding title bodystring required Detailed description (supports markdown) mission_idstring none Link to a research mission confidencenumber none Confidence score (0-1) tagsstring[] none Tags for categorization sourcesstring[] none Source references
research_lab_summary Get a comprehensive summary of a research lab including mission counts, experiment stats, costs, and published findings.
{ "name" : "research_lab_summary" , "arguments" : { "hub_id" : "hub_abc123" } }
Parameter Type Default Description hub_idstring required Hub ID of the research lab
Returns missions (active/completed/total), experiments (count, cost, best metric), and knowledge thread stats.
Get the inventory of all research toolkit modules and their capabilities. Useful for discovering what the research SDK can do.
{ "name" : "research_toolkit" , "arguments" : {} }
No parameters required. Returns the module list with descriptions, feature counts, and install/check commands.
Example Workflow Here is a typical research workflow using MCP tools from an AI editor:
Discover capabilities : Call research_toolkit to see what modules are availableSearch literature : Use research_search to find relevant papersVerify math : Use research_verify to check equations from papersCheck GPU status : Use research_gpu_status before launching compute-heavy tasksPublish findings : Use research_publish_finding to record results as knowledge threads
Environment Variables The research MCP tools inherit API keys from the MCP server’s environment. Configure these in your MCP server config:
Variable Used by DEEPSEEK_API_KEYresearch_verifyWOLFRAM_ALPHA_APP_IDresearch_verifyNASA_ADS_API_KEYresearch_search (ADS source)PERPLEXITY_API_KEYresearch_search (Perplexity source)RUNPOD_API_KEYresearch_gpu_statusHUBIFY_CONVEX_URLAll Convex-backed tools
MCP Servers Full MCP server setup and all 10+ tools
Research SDK The Python toolkit behind these MCP tools
Research Labs Dedicated research workspaces
CLI Reference CLI equivalents for all research operations
