Open Design Intelligence

Open-source MCP server that gives AI coding agents access to 1,320 reviewed design references. No database, model API, or runtime inference required.

Hosted MCP

Connect directly — no install or credentials needed:

https://open-design-intelligence.vercel.app/mcp

Claude Code:

claude mcp add --transport http open-design-intelligence https://open-design-intelligence.vercel.app/mcp

Codex (~/.codex/config.toml):

[mcp_servers.open-design-intelligence]
url = "https://open-design-intelligence.vercel.app/mcp"

Cursor and VS Code can connect to the same URL as a remote Streamable HTTP server. The website has copyable, client-specific instructions.

Run locally from npm

Requires Node.js 22 or newer. No global installation is needed:

npx -y open-design-intelligence@1.0.0

The command starts an MCP stdio server, so it normally waits silently for an MCP client rather than opening an interactive prompt. Configure it in a client with:

{
  "mcpServers": {
    "open-design-intelligence": {
      "command": "npx",
      "args": ["-y", "open-design-intelligence@1.0.0"]
    }
  }
}

Pin the exact version so a client update cannot silently change the server or corpus.

MCP tools

Tool	Purpose
search_designs	Search approved records using lexical, trait, quality, and taxonomy signals.
get_design	Retrieve selected canonical sections, Markdown, and metadata.
recommend_designs	Recommend references for a product brief or design query.
find_similar	Find deterministic similarity neighbours for one design.
find_patterns	Find recurring decisions, variants, and counterexamples.
design_consensus	Answer a question across a bounded corpus sample without overstating consensus.
design_synthesis	Build a canonical design brief with evidence mappings and explicit assumptions.
design_critique	Audit user-provided DESIGN.md Markdown for specificity and implementation usefulness.
explain_design	Answer a question about one design with supporting, contradicting, and missing evidence.

Evidence and limitations

Search combines deterministic lexical ranking with local sparse-vector cosine scoring — no embedding APIs. Outputs distinguish direct record evidence, derived corpus observations, and assumptions. Missing data is not negative evidence. The MCP is read-only.

Repository

apps/vercel       Vercel deployment and hosted MCP adapter
apps/web          Static public website
designs           Approved canonical design records
generated         Deterministic runtime artifacts
packages/cli      Corpus validation and index generation
packages/intelligence  Evidence-backed analysis
packages/mcp      stdio and Streamable HTTP MCP server
packages/parser   Canonical Markdown parser and validator
packages/schemas  Shared public contracts
packages/search   Prepared retrieval and ranking

Run the full verification suite:

pnpm verify

Contribution requirements are in CONTRIBUTING.md. Report vulnerabilities per SECURITY.md. Licensed under MIT.