streamfind

streamfind is an open-source platform for reproducible analytical data processing. Projects are stored in DuckDB and processing is expressed as validated workflows. The platform includes a browser application, native C++ runtime, MCP interface, and extensible domain plugins.

Start here

• Download the current native packages.
• Launch and develop the browser application.
• Develop backend and frontend plugins.
• Use the C++ MCP server.
• Check supported interfaces and limitations.
• Use the R interface.

For an overview of the system boundaries and data flow, see the architecture page. For reproducible native validation, see the testing guide.

Typical workflow

1. Install a native package and launch the browser application, or configure an MCP client.
2. Create or open a DuckDB project.
3. Discover operations and configure their parameters.
4. Connect compatible typed ports into a workflow.
5. Validate and run the workflow.
6. Inspect tables, visualizations, results, and execution history.

Version 0.7.1

The native C++ package version is 0.7.1. Packages target Windows x64 and Linux x86_64. Cross-version C++ ABI stability is not guaranteed; use matching runtime and developer-kit versions.

Interfaces

Interface	Provides	Availability
C++ core	Native project API, mass-spectrometry operations, and MCP server	Available as a native package
MCP	JSON-RPC over stdio for applications and AI agents	Available through the native C++ package
R package	Existing R workflows, non-target screening, and Shiny application	Separate R interface
Python package	—	No public package
React web app	Browser application served by the packaged C++ service; also available from the Vite development server	Included in the Windows and Linux native archives

Supported analytical work

The mass-spec plugin supports workflows that combine analysis metadata, raw file readers, spectra, chromatograms, feature detection and grouping, isotope and molecular-link annotation, suspect screening, and downstream filtering or visualization. Exact reader and operation support depends on the file format and operation parameters.

MCP at a glance

The C++ MCP server exposes the catalogue-backed interface.

1. Call initialize to receive the server capabilities and usage guidance.
2. Call tools/list to discover callable Operations.
3. Use create, then describe, get_domain, or get_metadata for a new project.
4. Use domain Operations with explicit database_path.
5. Add operation nodes with their parameters to the persisted workflow.
6. Connect typed output ports to downstream input ports.
7. Validate the graph, run it, and inspect the resulting published results.

Operations are the units of execution. A workflow is a persisted graph of operation nodes and typed connections. See the C++ MCP quickstart for request examples.

Shared contract

The semantic catalogue defines operation names, parameters, nested input schemas, results, units, constraints, and agent-facing guidance. The C++ backend implements the contract and exposes the public concepts through its API and MCP server.

The architecture page explains the relationship between the catalogue, native APIs, and MCP. The current React application uses the C++ backend through its public service boundary; it does not access project DuckDB files directly. See Web app for packaged Windows and Linux launch instructions and development setup.