You.com MCP Server — you-aware

A Model Context Protocol (MCP) server that turns the context your agent harness already maintains (CLAUDE.md, Cursor rules, Cline context) into smarter retrieval, backed by the You.com Search API. The harness owns the memory; you-aware owns the retrieval intelligence layer that uses it. Built with the Bun runtime and supports multiple transport protocols.

Features

• Context-aware retrieval: Pass harness context as parameters — trusted_sources, blocked_sources, project_context, prior_decisions, workflow_stage — and the server applies it via source boosts/blocks, query planning, freshness tuning, and ranking.

• Opinionated technical-research profile: Sensible defaults for developer research (boosted docs domains, blocked low-quality domains, stage-aware freshness and ranking) — configurable at the edges.

• Multi-query planning: Decomposes a request into parallel sub-queries (base, trusted-source, subtopic, prior-decision, context-expansion), then merges, dedupes, and reranks into one logical search.

• Local retrieval cache: Per-sub-query disk cache (~/.you-aware/cache) for fast cross-session repeats. Stdio mode only.

• Decisions ledger: you-record-decision appends research decisions to ./.you-aware/decisions.jsonl; future searches read them back to auto-populate prior_decisions. Stdio mode only.

• Inspectable traces: Every response carries a JSON trace (profile, resolved params, planned sub-queries, cache hits/misses, dedup counts, ranking decisions).

• Multiple transport protocols: Stdio (local, full local-state) and Streamable HTTP (per-request Bearer token, local state disabled).

• TypeScript + Zod: Full type safety and runtime validation.

Related MCP server: mcp-omnisearch

Adding to your MCP client

This server can be integrated with MCP clients in two ways:

• Option 1: Remote Server (Recommended) - No installation required, uses hosted server at https://api.you.com/mcp with HTTP transport and API key authentication

• Option 2: Local NPM Package - Install via npx @youdotcom-oss/mcp with stdio transport, environment variable authentication, and runs locally on your machine

Standard Configuration Templates

Remote Server (Recommended):

{
 "mcpServers": {
 "ydc-search": {
 "type": "http",
 "url": "https://api.you.com/mcp",
 "headers": {
 "Authorization": "Bearer <you-api-key>"
 }
 }
 }
}

Local NPM Package:

{
 "mcpServers": {
 "ydc-search": {
 "command": "npx",
 "args": ["@youdotcom-oss/mcp"],
 "env": {
 "YDC_API_KEY": "<you-api-key>"
 }
 }
 }
}

Quick Setup:

# Add using Claude Code CLI (if available)
claude mcp add ydc-search npx @youdotcom-oss/mcp

Manual Setup:

1. Follow the Claude Code setup guide

2. Create or update .mcp.json in your workspace root using the standard configuration template above

3. For remote server: add "type": "http" to the configuration

4. For local package: add "type": "stdio" to the configuration

Setup: Use the standard configuration template above in your Claude Desktop MCP configuration.

Installation: Follow the Claude Desktop MCP guide for setup.

Download Claude Desktop

Setup: Edit ~/.codex/config.toml:

[mcp_servers.ydc-search]
command = "npx"
args = ["@youdotcom-oss/mcp"]

[mcp_servers.ydc-search.env]
YDC_API_KEY = "<you-api-key>"

Download Codex

GUI Setup (Easiest):

1. Go to Cursor Settings > Features > MCP

2. Click "+ Add New MCP Server"

3. For remote: Select "Streamable HTTP" transport, URL: https://api.you.com/mcp

4. For local: Select "stdio" transport, Command: npx, Args: @youdotcom-oss/mcp

Manual Setup: Create .cursor/mcp.json in your project directory or ~/.cursor/mcp.json globally using the standard configuration template above.

Note: Remove the "type" field from the remote server configuration for Cursor.

Documentation | Download Cursor

Setup: Use the standard configuration template above in your Gemini CLI MCP server configuration.

Installation:

1. Install Gemini CLI

2. Follow the MCP server setup guide

Documentation | Download Gemini CLI

Quick Setup: Go to "Advanced settings" → "Extensions" → "Add custom extension"

Manual Setup: Use the standard configuration template above in your Goose extensions configuration.

Installation Guide | Download Goose

Setup: Configure in your IDE settings using the local NPM package configuration from the standard template above.

For Remote Server: Use mcp-remote since JetBrains only supports stdio transport:

{
 "mcpServers": {
 "ydc-search": {
 "command": "npx",
 "args": ["mcp-remote", "https://api.you.com/mcp", "--header", "Authorization: Bearer ${YDC_API_KEY}"],
 "env": { "YDC_API_KEY": "<you-api-key>" }
 }
 }
}

Supported IDEs: IntelliJ IDEA, PyCharm, WebStorm, etc. (requires AI Assistant enabled)

Documentation

Setup: Edit mcp.json in LM Studio settings using the standard configuration template above.

Installation: Configure through program settings or edit configuration file manually.

Download LM Studio

Setup: Edit ~/.config/opencode/opencode.json:

{
 "$schema": "https://opencode.ai/config.json",
 "mcp": {
 "ydc-search": {
 "type": "local",
 "command": ["npx", "@youdotcom-oss/mcp"],
 "enabled": true,
 "env": { "YDC_API_KEY": "<you-api-key>" }
 }
 }
}

For Remote Server:

{
 "$schema": "https://opencode.ai/config.json",
 "mcp": {
 "ydc-search": {
 "type": "local",
 "command": ["npx", "mcp-remote", "https://api.you.com/mcp", "--header", "Authorization: Bearer ${YDC_API_KEY}"],
 "enabled": true,
 "env": { "YDC_API_KEY": "<you-api-key>" }
 }
 }
}

Documentation

Setup:

1. Open Qodo Gen chat panel in VSCode or IntelliJ

2. Click "Connect more tools" → "+ Add new MCP"

3. Paste the standard config above

4. Click Save

Documentation

Quick Setup (Command Line):

# Add MCP server
code --add-mcp "{\"name\":\"ydc-search\",\"command\":\"npx\",\"args\":[\"@youdotcom-oss/mcp\"],\"env\":{\"YDC_API_KEY\":\"<you-api-key>\"}}"

Manual Setup: Create mcp.json file in your workspace (.vscode/mcp.json) or user profile using the standard configuration template above, but replace "mcpServers" with "servers".

Secure Setup with Input Prompts:

{
 "inputs": [
 {
 "type": "promptString",
 "id": "ydc-api-key",
 "description": "You.com API Key",
 "password": true
 }
 ],
 "servers": {
 "ydc-search": {
 "command": "npx",
 "args": ["@youdotcom-oss/mcp"],
 "env": { "YDC_API_KEY": "${input:ydc-api-key}" }
 }
 }
}

Requirements: GitHub Copilot extension must be installed

Documentation | Download VS Code

Setup: Use the standard configuration template above.

Installation: Follow MCP documentation for Windsurf-specific setup instructions.

Documentation | Download Windsurf

Setup: Add to your Zed settings.json using "context_servers" instead of "mcpServers":

{
 "context_servers": {
 "ydc-search": {
 "source": "custom",
 "command": "npx",
 "args": ["@youdotcom-oss/mcp"],
 "env": {
 "YDC_API_KEY": "<you-api-key>"
 }
 }
 }
}

For Remote Server: Use mcp-remote to bridge HTTP to stdio:

{
 "context_servers": {
 "ydc-search": {
 "source": "custom",
 "command": "npx",
 "args": ["mcp-remote", "https://api.you.com/mcp", "--header", "Authorization: Bearer ${YDC_API_KEY}"],
 "env": { "YDC_API_KEY": "<you-api-key>" }
 }
 }
}

Setup Instructions | Download Zed

General Configuration Notes

• Remote Server: Recommended for most users - no installation required, just API key

• NPM Package: Alternative for local usage or when you prefer running locally

• HTTP Transport: Use for remote server connections and web applications

• Stdio Transport: Use for local npm package installations and development

• API Key: Always required - either as environment variable (stdio) or in headers (http)

• Docker/Local Development: See sections below for advanced local development setups

See the Transport Protocols section for detailed protocol information.

Building and Running Locally

Prerequisites

• Bun 1.2.21 or higher (replaces Node.js)

• You.com API key (get one at api.you.com)

Local Workspace Setup

Since this package is not published to npm (marked as private), you need to clone and set it up locally:

# Clone the repository
git clone <repository-url>
cd you-mcp-server

# Install dependencies
bun install

# Set up your environment file with your You.com API key (optional)
echo "export YDC_API_KEY=<you-api-key>" > .env

Building

# Build is optional for development, required for production bin executables
bun run build # Builds only stdio.ts to dist/stdio.js

For MCP Client Integration: Use the full path to your local server installation in your .mcp.json:

{
 "mcpServers": {
 "ydc-search": {
 "type": "stdio",
 "command": "bun",
 "args": ["/full/path/to/you-mcp-server/src/stdio.ts"],
 "env": {
 "YDC_API_KEY": "<you-api-key>"
 }
 }
 }
}

Alternative using built executable:

{
 "mcpServers": {
 "ydc-search": {
 "type": "stdio",
 "command": "node",
 "args": ["/full/path/to/you-mcp-server/bin/stdio"],
 "env": {
 "YDC_API_KEY": "<you-api-key>"
 }
 }
 }
}

Configuration

Set up your environment file with your You.com API key:

echo "export YDC_API_KEY=<you-api-key>" > .env
source .env

Replace <you-api-key> with your actual API key:

echo "export YDC_API_KEY=your-actual-api-key-here" > .env
source .env

Alternatively, set it as an environment variable:

export YDC_API_KEY="your-api-key-here"

Available Scripts

• bun run dev - Start server in stdio mode for development

• bun run build - Build stdio.ts to dist/ for production

• bun start - Start HTTP server on port 4000 (or PORT env var)

• bun run test - Run test suite

• bun run check - Run Biome linting and formatting checks

Executable Scripts

The project includes executable scripts in bin/:

• ./bin/stdio - Stdio transport server (requires bun run build first)

• ./bin/http - HTTP transport server (runs directly from source)

Running the Server

Stdio Mode (Recommended for MCP Clients) - For local workspace integration:

# Development mode (direct source)
bun run dev
# or
bun src/stdio.ts

# Production mode (built distribution)
bun run build # Build first
./bin/stdio # Run built version

HTTP Mode - For web applications and remote clients:

# Default port 4000
bun start
# or
./bin/http

# Custom port
PORT=8080 bun start
# or
PORT=8080 ./bin/http

Docker Deployment

Build and run with Docker:

# Build the optimized Docker image (243MB final size)
docker build -t youdotcom-mcp-server .

# Run the container
docker run -d -p 4000:4000 --name youdotcom-mcp youdotcom-mcp-server

Optimization Features:

• Multi-stage build: Uses standalone binary compilation with bun build --compile

• Minimal base image: Ubuntu 22.04 with no additional packages

• Size optimized: 243MB final image

• Self-contained: Includes Bun runtime in compiled binary

• Security: Runs as non-root user, minimal attack surface

Using Docker Compose:

Create a docker-compose.yml file:

version: '3.8'
services:
 you-mcp-server:
 build: .
 ports:
 - "4000:4000"
 environment:
 - YDC_API_KEY=${YDC_API_KEY}
 - PORT=4000
 restart: unless-stopped

Then run:

docker-compose up -d

Claude Code Setup with Docker

To use this MCP server with Claude Code via Docker:

1. Start the Docker container:

   docker run -d -p 4000:4000 --name youdotcom-mcp youdotcom-mcp-server

2. Configure Claude Code:

   • Copy .mcp.example.json to .mcp.json

   • Replace <you.com api key> with your actual You.com API key

   cp .mcp.example.json .mcp.json

   Your .mcp.json should look like:

   {
    "mcpServers": {
    "ydc-search": {
    "type": "http",
    "url": "http://localhost:4000/mcp",
    "headers": {
    "Authorization": "Bearer <you-api-key>"
    }
    }
    }
   }

3. Verify the setup:

   • The server will be available at http://localhost:4000/mcp

   • Health check endpoint: http://localhost:4000/mcp-health

API Reference

you-aware

Context-aware web and news search. Applies the technical-research profile, honours harness context, plans parallel sub-queries, caches locally (stdio mode), and returns results with an inspectable retrieval trace.

Harness-context parameters:

• trusted_sources (string[], optional): Domains to boost in ranking and plan dedicated sub-queries for (e.g. ["docs.python.org", "github.com"]). Max 20.

• blocked_sources (string[], optional): Domains to exclude. Applied as -site: operators and as a post-fetch filter. Max 50.

• project_context (string, optional): Free-text project facts (stack, constraints, goals) from the harness. Seeds query expansion. Max 8000 chars.

• prior_decisions (string[] | {decision, tags?}[], optional): Prior research decisions to bias retrieval. Merged with decisions auto-read from the local ledger. Max 20.

• workflow_stage (string, optional): exploring | implementing | debugging | reviewing. Tunes freshness, result count, and ranking weights (e.g. debugging favours recency).

• max_subqueries (integer, optional): Cap on planned parallel sub-queries. Default 3, hard max 5.

Base search parameters (same as the underlying You.com Search API):

• query (string, required): The base search query. Supports operators directly: + (exact term), - (exclude term), site:, filetype:, lang:. Use parentheses for multi-word phrases.

• site (string, optional): Search within a specific website domain (e.g., 'github.com')

• fileType (string, optional): Filter by a specific file type (e.g., 'pdf', 'doc', 'txt')

• language (string, optional): Filter by a specific language using ISO 639-1 code (e.g., 'en', 'es', 'fr')

• exactTerms (string, optional): Exact terms with logical operators: 'python AND|tutorial|NOT beginner' (pipe-separated). Use parentheses for multi-word phrases.

• excludeTerms (string, optional): Terms to exclude with logical operators (pipe-separated). Cannot be used with exactTerms.

• count (integer, optional): Maximum number of results to return per section. Range: 1-20.

• freshness (string, optional): Freshness of results. Options: day, week, month, year.

• offset (integer, optional): Offset for pagination (calculated in multiples of count). Range: 0-9.

• country (string, optional): Country code for localized results. Examples: US, GB, DE, FR, JP, CA, AU, etc.

• safesearch (string, optional): Content filtering level. Options: off, moderate (default), strict.

Returns:

The same content text body as before, plus a trace object in structuredContent describing the profile applied, resolved params, planned/executed sub-queries, cache hits/misses, dedup counts, and ranking decisions. Example (abbreviated):

{
 "content": [
 {
 "type": "text",
 "text": "Search Results for \"machine learning\":\n\nWEB RESULTS:\n\nTitle: Introduction to Machine Learning\nURL: https://github.com/ml-tutorials/intro\nDescription: A comprehensive guide to machine learning fundamentals\nSnippets:\n- Learn the basics of supervised and unsupervised learning\n- Practical examples with Python and TensorFlow\n\n---\n\nTitle: Machine Learning Course\nURL: https://coursera.org/ml-course\nDescription: Stanford's machine learning course materials\nSnippets:\n- Mathematical foundations of ML algorithms\n- Hands-on programming assignments\n\n==================================================\n\nNEWS RESULTS:\n\nTitle: AI Breakthrough in Medical Diagnosis\nURL: https://techcrunch.com/ai-medical-breakthrough\nDescription: New machine learning model achieves 95% accuracy\nPublished: 2024-01-15T10:30:00"
 }
 ],
 "structuredContent": {
 "results": {
 "web": [
 {
 "url": "https://github.com/ml-tutorials/intro",
 "title": "Introduction to Machine Learning",
 "description": "A comprehensive guide to machine learning fundamentals",
 "snippets": [
 "Learn the basics of supervised and unsupervised learning",
 "Practical examples with Python and TensorFlow"
 ],
 "page_age": "2024-01-10T14:20:00",
 "authors": ["ML Tutorial Team"]
 }
 ],
 "news": [
 {
 "url": "https://techcrunch.com/ai-medical-breakthrough",
 "title": "AI Breakthrough in Medical Diagnosis",
 "description": "New machine learning model achieves 95% accuracy",
 "page_age": "2024-01-15T10:30:00"
 }
 ]
 },
 "metadata": {
 "query": "machine learning",
 "request_uuid": "f47ac10b-58cc-4372-a567-0e02b2c3d479",
 "latency": 0.247
 },
 "trace": {
 "profile": "technical-research",
 "resolved_params": { "count": 10, "workflow_stage": "implementing" },
 "trusted_sources": ["github.com", "docs.python.org"],
 "blocked_sources": ["w3schools.com"],
 "prior_decisions_used": 1,
 "prior_decisions_source": { "explicit": 0, "ledger": 1 },
 "subqueries": [
 { "query": "machine learning", "origin": "base", "cache": "miss", "latency_ms": 210, "web_count": 10, "news_count": 2 },
 { "query": "machine learning site:github.com", "origin": "trusted_source", "cache": "hit", "latency_ms": 1, "web_count": 10, "news_count": 0 }
 ],
 "cache": { "hits": 1, "misses": 1, "key": "1a2b3c" },
 "dedup": { "web_before": 20, "web_after": 14, "news_before": 2, "news_after": 2 },
 "local_state_enabled": true,
 "total_latency_ms": 247
 }
 }
}

you-record-decision

Appends a research-driven decision to the project-local ledger so future you-aware searches can build on it. Stdio mode only — in HTTP mode this is disabled and returns { "recorded": false }.

Parameters:

• decision (string, required): The decision that was made (e.g. "Use Bun.serve over Express").

• rationale (string, optional): Why the decision was made.

• query (string, optional): The search query that prompted the decision.

• sources (string[], optional): URLs or domains the decision was based on.

• tags (string[], optional): Topic tags used to relate the decision to future searches.

Returns: { "recorded": true, "timestamp": "<ISO-8601>" } on success.

Local state and storage

Filesystem features are active only in stdio mode (local developer machine). In HTTP mode (per-request, potentially multi-tenant) they are disabled by construction.

• Cache — ~/.you-aware/cache (global, cross-session). Override with the YOU_AWARE_CACHE_DIR environment variable.

• Decisions ledger — ./.you-aware/decisions.jsonl (project working directory). Override with the YOU_AWARE_LEDGER_DIR environment variable.

We never store your harness context; only the local cache entries and the decisions you explicitly record are written to disk.

Maintenance

Resources

• NPM Package
• GitHub Repository
Need Help?Related Servers