@hoklims/techspace-mcp

Pareto-optimal tech stack recommendations for AI coding agents. MCP server exposing a deterministic engine over 125 technologies and 9 scoring dimensions. No LLM in the scoring path.

Install

Claude Code, Cursor, Cline, Continue, Codex, or any other MCP client

Add to your MCP configuration:

{
	"mcpServers": {
		"techspace": {
			"command": "npx",
			"args": ["-y", "@hoklims/techspace-mcp@latest"]
		}
	}
}

Claude Desktop (one-click)

Download the latest techspace-mcp.mcpb from the releases page, then drag-drop into Claude Desktop → Settings → Extensions → Advanced → Install Extensions.

Claude Desktop ships its own Node.js runtime, so no additional installation is required.

Manual install

npm install -g @hoklims/techspace-mcp
techspace-mcp

The package name on npm is @hoklims/techspace-mcp (scoped). The command it installs is techspace-mcp (unscoped). The binary reads JSON-RPC from stdin and writes to stdout.

Tools

| Tool | Purpose | |---|---| | explore_tradeoffs | Main entry. Generates a Pareto front of optimal stacks given project constraints. | | adjust_preference | Shifts the Pareto front along one axis (more performance, simpler, cheaper). | | explain_integration | Details on a specific cross-ecosystem connector. | | get_technology | Deep-dive on a single technology (scores, relations, deprecation). | | score_stack_compatibility | Compatibility score for a custom stack composition. |

Example

User: I am building a self-hosted SaaS with multi-tenant auth and vector search. Team of 3. Prefer JavaScript.

Agent: calls explore_tradeoffs({ required_categories: ['meta_framework', 'database', 'auth'], preferred_ecosystem: 'javascript', needs_vector_db: true, self_hosted: true, team_size: 'small' }) and receives four Pareto-optimal stacks, each with integration connectors and tradeoffs.

Environment variables

• TECHSPACE_SOLVER_URL (optional) — URL of a running TechSpace CP-SAT solver (Python microservice) for exact Pareto optimization. When unset, a fast greedy fallback is used (sub-100 ms). Power users only.

Why deterministic

The scoring path uses zero LLM. Technologies, relations, and scores are curated JSON in the source repository. The Pareto solver is either CP-SAT (OR-Tools, remote) or a greedy fallback (bundled). The LLM's role is only to extract constraints from the conversation and call the right tool.

Result: recommendations are reproducible. Same input, same output, every time.

Source

github.com/hoklims/techspace — MIT license.