Agents do not fail because the model is weak. They fail because the knowledge layer underneath them is four services glued together: a vector database, an embedding pipeline, a sync job, and a hand-written MCP server that drifts out of sync with all three.
FoxNose is one managed service that replaces that stack. REST and MCP from the same endpoint, over the same data, with the same schema.
- Hybrid search. Vector similarity, full-text matching and structured filters resolve in a single query, not three you merge yourself.
- Auto-embeddings. Documents are embedded on write. There is no pipeline to run and nothing to backfill when a document changes.
- Knowledge MCP. Every collection gets a generated MCP server with a fixed tool surface and JSON Schema introspection, so an agent discovers the shape of your data instead of being told about it in a prompt.
- Schema versioning. Pin an agent to a schema version and migrate the collection without breaking it mid-flight.
- Multi-tenant by default. Per-key scopes, collection isolation, RBAC and 30-day immutable audit logs.
Works with OpenAI, Anthropic and Google models, with LangChain and Ollama, and with MCP clients including Claude Code, Claude Desktop, Cursor, Cline and Continue.
| Repository | What it is |
|---|---|
| foxnose-python | Official Python SDK. On PyPI as foxnose-sdk. |
| foxnose-typescript | Official TypeScript SDK. |
| langchain-foxnose | LangChain integration for Python: retriever and vector store. |
| langchain-foxnose-js | LangChain integration for JavaScript and TypeScript. |
- foxnose.net — what it is and who it is for
- Documentation — quickstart, API reference, MCP setup
- Pricing — free tier, no card
- Blog — how the thing is built
Questions, bug reports and integration requests: open an issue on the relevant repository, or write to support@foxnose.net.