feat: add OpenRouter as embedding provider name

[BeamNawapat]

Summary

Add OpenRouter as a first-class provider name for cleaner UX. Reuses OpenAIEmbedding internally with OpenRouter's OpenAI-compatible endpoint — zero code duplication.

Motivation

Users can achieve the same result via OPENAI_BASE_URL workaround, but a named provider gives better discoverability and simpler configuration.

Changes (2 files, ~30 lines)

• packages/mcp/src/config.ts — Add OpenRouter to provider union, OPENROUTER_API_KEY env var, default model, help message
• packages/mcp/src/embedding.ts — Add OpenRouter case that creates OpenAIEmbedding with baseURL: 'https://openrouter.ai/api/v1'

No new files. No new dependencies. No code duplication.

Usage

EMBEDDING_PROVIDER=OpenRouter \
OPENROUTER_API_KEY=sk-or-v1-xxx \
EMBEDDING_MODEL=openai/text-embedding-3-small \
npx @zilliz/claude-context-mcp@latest

Test plan

• OpenRouter provider starts and connects
• Existing providers unaffected (no regression)
• pnpm build passes

[Copilot]

Pull request overview

Adds OpenRouter as a first-class embedding provider (via OpenRouter’s OpenAI-compatible API) so users no longer need to rely on the OPENAI_BASE_URL workaround and can use OpenRouter-style model IDs like openai/text-embedding-3-small.

Changes:

• Introduce OpenRouterEmbedding provider in core (OpenAI SDK + OpenRouter base URL, supported-model metadata).
• Wire OpenRouter into MCP config/env (OPENROUTER_API_KEY) and embedding-provider selection switch.
• Export the new provider from the core embedding barrel.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
packages/core/src/embedding/openrouter-embedding.ts	New embedding provider implementation using OpenAI SDK routed through OpenRouter.
packages/core/src/embedding/index.ts	Exports the new OpenRouter embedding provider.
packages/mcp/src/config.ts	Extends config/provider union, adds OpenRouter default model and API key env wiring.
packages/mcp/src/embedding.ts	Adds OpenRouter provider case to instance creation and provider-info logging.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

OpenRouterEmbedding.getDimension() returns the default this.dimension for unknown/custom models without any warning. Since other implementations (e.g., OpenAIEmbedding) explicitly warn that the value may be incorrect until detectDimension()/embed() is called, this can lead to misleading logs and harder debugging when users supply non-listed OpenRouter model IDs. Consider adding a warning for the unknown-model path (or otherwise documenting that the returned dimension may be stale until detection runs).

Suggested change

	console.warn(
	`OpenRouter model "${model}" is not in the supported models list; returning cached/default dimension ` +
	`${this.dimension}. This value may be incorrect until detectDimension(), embed(), or embedBatch() runs.`
	);

[Copilot]

createMcpConfig() prints a debug summary of relevant env vars (e.g., GEMINI_API_KEY, OPENAI_API_KEY), but the newly introduced OPENROUTER_API_KEY isn’t included. Adding it to the debug output (ideally with the same “SET (length: …)” pattern) would make troubleshooting OpenRouter configuration consistent with the other providers.

[zc277584121]

Thanks for the contribution! The motivation makes sense — giving OpenRouter users a cleaner config experience. However, I have some concerns about the current approach.

Main issue: code duplication

openrouter-embedding.ts (151 lines) is nearly identical to openai-embedding.ts. Both use the openai npm package, same embeddings.create() calls, same error handling, same detectDimension logic. The only real differences are:

• Default baseURL → https://openrouter.ai/api/v1
• Default model → openai/text-embedding-3-small
• Supported model list (3 entries, all just OpenAI models with openai/ prefix)

Users can already achieve the same result today:

EMBEDDING_PROVIDER=OpenAI \
OPENAI_BASE_URL=https://openrouter.ai/api/v1 \
OPENAI_API_KEY=sk-or-v1-xxx \
EMBEDDING_MODEL=openai/text-embedding-3-small

Suggested alternatives

If we do want a first-class OpenRouter provider name for better UX, it should reuse OpenAIEmbedding rather than duplicate it. Something like:

// In createEmbeddingInstance:
case 'OpenRouter':
    if (!config.openrouterApiKey) {
        throw new Error('OPENROUTER_API_KEY is required');
    }
    return new OpenAIEmbedding({
        apiKey: config.openrouterApiKey,
        model: config.embeddingModel,
        baseURL: 'https://openrouter.ai/api/v1',
    });

This achieves the same user-facing behavior (~5 lines vs 151 lines) and avoids maintaining two copies of the same logic.

Alternatively, just document the OPENAI_BASE_URL approach in the README — that may be sufficient.

[BeamNawapat]

Good call — 151 lines of duplication was unnecessary. Reworked to reuse OpenAIEmbedding directly:

case 'OpenRouter':
    return new OpenAIEmbedding({
        apiKey: config.openrouterApiKey,
        model: config.embeddingModel,
        baseURL: 'https://openrouter.ai/api/v1',
    });

Force-pushed. Now 2 files changed, ~30 lines total. No new files, no new dependencies.