Command Palette

Search for a command to run...

0

Selected by You.com: How a Bot PR Became an Architecture Validation

You.com scanned 93 OSS AI agent projects to promote their search API. agent-search-mcp won the MCP search category. How a growth-hack PR turned into an unexpected architecture endorsement and market positioning report.

Selected by You.com: How a Bot PR Became an Architecture Validation

A promotional bot PR taught me more about my search engine's market position than any user feedback.


1. A Strange PR Arrives

On July 23, agent-search-mcp received an external PR: "feat: add optional You.com search engine." Code quality was solid — 67-line engine adapter, 4 tests, follows conventions. Looks like a standard community contribution.

But a few details felt off:

  • Author mouse-value-add: created Feb 2026, 64 public repos, 0 followers, bio "friendly, small"
  • PR description follows a rigid template (Problem → Solution → Setup → Validation)
  • Bottom of the PR links to a tracking issue: youdotcom-oss/integration-tracking/issues/116

I clicked through. That's when the full picture emerged.


2. You.com's Industrial-Scale OSS Promotion

What You.com Is

You.com is an AI search engine with a search API. Like Tavily, Exa, and Brave, they want AI agents using their API for web searches.

93 Issues, One Playbook

The youdotcom-oss/integration-tracking repo contains 93 open issues — all tracking the same thing: PRs submitted to OSS projects to add You.com as a search provider.

Every issue follows the exact same structure:

[Integration] repo-name — integration approach

### Candidate Evaluation
- ProjectA: SELECTED — best fit
- ProjectB: too coupled
- ProjectC: too large

### Integration Plan
- Optional engine adapter
- Default-off
- Selectable via --engines

### Current Status
PR opened, awaiting review

This isn't random outreach. It's a methodical, prioritized growth engine — with candidate evaluation, risk assessment, and progress tracking.

The Four-Tier Target Funnel

The 93 tracked projects weren't chosen at random. By ecosystem influence, they form four clear tiers:

TierPositionExamples
SPlatform/framework — integration reaches all downstream usersmicrosoft/semantic-kernel, deepset-ai/haystack, Alibaba-DeepResearch
AWell-known OSS — large independent user basesHermes Agent, meilisearch, Perplexica
BCategory leader — occupies a distribution node in a verticalagent-search-mcp, mcp-omnisearch, swarmclaw
CPersonal/small — stars < 50Numerous personal repos

As of this investigation:

  • Tier C: 10+ merged (small maintainers merge on sight)
  • Tier B: Most PRs still open (maintainers reviewing)
  • Tier S: No PRs submitted yet (evaluating, or hesitant)

Zero S or A-tier projects have merged. Big projects treat bot-PRs very differently than small ones.


3. Why We Were Selected

In the MCP search vertical, You.com evaluated three projects side-by-side:

ProjectStarsResultReasoning
agent-search-mcp9✅ SELECTED"clean engine registry, tool-level engine selection, test coverage — best overall fit"
n24q02m/wet-mcp15Stronger search stack but tightly coupled to SearXNG/local crawling
tobocop2/lilbee38Repo too large, platform-oriented, PR impact uncontrollable

A 9-star project beat a 38-star competitor. We didn't win on popularity — we won on architecture:

  1. Clean engine registry — adding an engine requires changes in 3 places
  2. Tool-level engine selection--engines / MCP params natively support optional engines
  3. Test coverage — engine parsing, policy filtering, search pipeline all tested, making changes low-risk

These are design principles we've held since Day 1 — a pluggable engine architecture. We designed it this way because we wanted to be the anti-Tavily: no vendor lock-in, every engine is optional. We didn't realize it would become our biggest competitive advantage for third-party integration.


4. The Search MCP Landscape

This investigation unexpectedly produced a competitive map of the "Agent Search Entry" category. The B-tier players:

ProjectStarsDifferentiator
mcp-omnisearch334Multi-engine aggregation, Tavily/Brave/Kagi/Exa
agent-search-mcp9Waterfall search, confidence scoring, progressive disclosure, Chinese optimization
swarmclaw624Agent runtime, search is one of many features

mcp-omnisearch's You.com PR was already merged — it's our closest competitor at 334 stars. But its differentiation is engine count (more API integrations). Ours is search quality optimization — waterfall search saves 50-75% API calls, progressive disclosure saves 36-58% tokens, semantic dedup/rerank adds precision.

Engine count can be copied. Structured search quality optimization is a moat.


5. Lessons for OSS Projects

This experience taught me several things about open-source positioning:

1. Architecture is Silent Marketing

Nobody promotes "my engine architecture is pluggable." But when third parties choose integration targets, this becomes the deciding factor. Good architecture sells itself.

2. Getting Bot-PRed Is a Category Signal

You.com doesn't PR random projects. Being selected means you're already a category entry point on their market map. Getting targeted ≈ being recognized.

3. Stars ≠ Distribution Value

We have 9 stars. A competitor has 334. You.com chose us. Because for them, "the likelihood a user will see You.com" depends on integration depth, not project fame.

4. Turn Competition Into Content

If I had just merged PR #15, this would be one more commit. Instead, I wrote this article — transforming a growth hack into third-party validation of our architecture.


6. What We Do Next

This validates two directions:

Double down on architecture:

  • Keep the engine registry pluggable (resist scope creep)
  • Every new feature adds ≤ 3 touch points
  • Maintain 480+ test density

Close the distribution gap:

  • The star gap (9 vs 334) means visibility is low
  • Add mcp.so, Smithery MCP directories
  • Chinese developer content (Juejin, V2EX)
  • Use "selected by You.com" as a README trust signal

This article itself is the strategy in practice — turning an accidental discovery into a brand story.

PR: github.com/lennney/agent-search-mcp/pull/15