Most search connectors are read-only; this one indexes too, which means an assistant can add what it found to the corpus it searches. That also raises the stakes on the key you give it — deleting the index is one of the operations, so a scoped key is the right call for anything shared. The index name is fixed in the environment, so one server instance serves one index.
A TypeScript MCP server over Azure Cognitive Search. It covers both directions: querying an index with customizable parameters, and putting documents into one — so an assistant can search a corpus and add to it in the same session.
- Full-text and semantic search over structured and unstructured data in your index, with a result count you control
- Document indexing — hand it records with their fields and they go into the Azure Cognitive Search index
- Index management, including deleting the current index
- Type-safe operations throughout, since the server is written in TypeScript
An Azure Cognitive Search service and three environment variables: `AZURE_SEARCH_ENDPOINT`, `AZURE_SEARCH_API_KEY` and `AZURE_SEARCH_INDEX_NAME`. Node.js v20 or higher. Run it with `npx -y hermes-search-mcp@latest`, or install it into Cline from the marketplace. The API key needs rights for what you intend to do — search-only keys will not index. The package is `hermes-search-mcp` (1.0.1 in package.json).
One command plus a key — npx -y hermes-search-mcp@latest, then supply credentials