createNamespace takes file storage, vector storage and embedding model configuration together, which puts those choices before the first document rather than after it — and resyncDocuments exists for when they change, naming a schema change as the reason to reprocess. The connector path has a consequence worth reading first: ingestConnector works from the document ids already in the backlog because the user picked them during authorisation, and when they are not there the user has to go through the authorisation flow again to pick them. hybridSearch requires an explicit hybridConfig with weights for keyword and semantic matching, so the balance between exact terms and meaning is a decision you make rather than a default you inherit.
A retrieval-augmented ingestion and search client with 25 tools covering namespaces, ingestion from many sources, document management and resync, semantic and hybrid search, and third-party connection lifecycle.
- Namespaces as the unit of configuration: createNamespace takes a file storage config, a vector storage config, an embedding model config and a web scraper config, with getNamespace, listNamespaces, updateNamespace and deleteNamespace alongside, and validateApiKey checking the key by listing them.
- Ingestion from wherever the content lives: ingestText for raw text with metadata and chunk configuration, ingestFile with automatic parsing across formats, ingestUrls for a list, ingestSitemap with path filtering and link limits, ingestWebsite for a recursive crawl with depth control, and ingestConnector which processes every document in a connector's backlog or failed state without needing ids.
- Job status: getIngestJobRunStatus checks a previously submitted ingestion run.
- Documents after they land: fetchDocuments with filter criteria, pagination and property selection, updateDocuments for metadata across a filter, deleteDocuments for a filtered set, and resyncDocuments to reprocess after a schema change.
- Retrieval in two modes: semanticSearch matches on meaning with a topK and a scoreThreshold, and hybridSearch combines keyword and semantic matching and requires a hybridConfig carrying the weight given to each.
- Third-party sources: createConnection returns an authorization URL the user opens to pick documents, listConnections and getConnection read the connections back, updateConnection re-runs the picker, and revokeConnection removes the integration.
- fetchUrlContent fetches a URL's content, described as particularly useful for parsed text file URLs.
A SourceSync API key with the namespace and tenant it should act in: SOURCESYNC_API_KEY, SOURCESYNC_NAMESPACE_ID and SOURCESYNC_TENANT_ID. A namespace has to exist with its file storage, vector storage and embedding model configured before anything can be ingested into it.
One command plus a key — env SOURCESYNC_API_KEY=your_api_key npx -y sourcesyncai-mcp, then supply credentials