for data scientists & analysts

dsos

An MCP server that keeps your coding agent's queries, datasets, charts, and narratives organized as versioned, searchable artifacts — so your analysis stops scattering across notebooks, chat scrollback, and one-off scripts.

watch the full recording ↗

what it gives you

artifact management — every query, dataset, transform, chart, and narrative saved, versioned, and searchable through one schema.
lineage — parents and children tracked automatically on every run — no manual bookkeeping.
skills & templates — workflow instructions and chart/report styling, both customizable and versioned like everything else.
GUI — a read-only browser over the same store: sessions, tool-call feed, artifact gallery, lineage graph.

$ pip install "git+https://github.com/zixiaoshawnshi/DataScientist-OS.git@latest"
$ claude mcp add dsos -s user -e DSOS_DB_PATH="$HOME/.dsos/store.db" -- python -P -m dsos.mcp_server

$ pip install "git+https://github.com/zixiaoshawnshi/DataScientist-OS.git@latest"
$ pi install npm:pi-mcp-adapter
# then add to ~/.pi/agent/mcp.json (use your absolute home path)
{ "mcpServers": { "dsos": { "command": "python", "args": ["-P", "-m", "dsos.mcp_server"],
  "env": { "DSOS_DB_PATH": "/home/you/.dsos/store.db" } } } }

$ pip install "git+https://github.com/zixiaoshawnshi/DataScientist-OS.git@latest"
$ codex mcp add dsos --env DSOS_DB_PATH="$HOME/.dsos/store.db" -- python -P -m dsos.mcp_server

$ pip install "git+https://github.com/zixiaoshawnshi/DataScientist-OS.git@latest"
# then add to opencode.json (use your absolute home path)
{ "mcp": { "dsos": { "type": "local", "command": ["python", "-P", "-m", "dsos.mcp_server"],
  "environment": { "DSOS_DB_PATH": "/home/you/.dsos/store.db" } } } }

or, hand this to your agent

Please look at https://github.com/zixiaoshawnshi/DataScientist-OS and set up dsos as an MCP server for me.

everything is an artifact

what	A registered source of data — `source` captures url, method, and fetch time, not just the file itself.
made via	`save_artifact(type="dataset", source=...)`
example	A CSV pulled from a public API, registered once before anything queries it.

what	A SQL query run against a dataset — the query text and its result both become artifacts, automatically.
made via	`run_sql(...)`
example	"group by category, count rows" — reusable the next time the same question comes up.

what	A Python transformation of one or more prior artifacts, recorded with its code, inputs, and status.
made via	`run_python(...)`
example	Joining two datasets, or computing a derived column.

what	A rendered PNG, styled by whichever chart-style template is active.
made via	`run_python(output_type="chart", style=...)`
example	A bar chart that later embeds into a report unchanged.

what	Markdown that embeds other artifacts by reference — the composition layer everything else feeds into.
made via	`save_artifact(type="narrative", content="... {{artifact:42}} ...")`
example	A write-up that pulls in three prior charts and a table, then `publish_report` renders it to one self-contained HTML file.

what	Markdown workflow instructions the agent reads and follows — teaches process, not code.
made via	`save_skill(...)` — an edit is a new version of the same skill
example	"how to discover a new dataset," seeded into every fresh store.

what	A chart style or report layout — versioned like everything else, not a hardcoded default.
made via	`save_template(kind=..., base=...)`
example	A custom dark report layout your narratives render into.