The MCP server#
The refinery ships an MCP server — a Model Context Protocol endpoint that lets AI assistants and other MCP clients work directly against a refinery instance. Through it an assistant can explore the catalog, sample and analyse data, read and author formulas, and manage dashboards, figures and balances — all under the same authentication and role model as the rest of the refinery.
Note
This is a pro-only feature of the Timeseries Refinery. If interested, please visit: https://timeseries.pythonian.fr/#ouroffer.
What it exposes#
Its capabilities fall into a few families:
discovery — search the catalog, read metadata, follow dependencies;
data — sample recent values, compute statistics, detect regimes and anomalies;
formulas — read, expand, test and (with write access) author computed series and groups;
dashboards, figures and balances — browse and manage the visualization objects;
baskets — reusable saved queries
notes - a per-user knowledge base;
tasks — inspect the scheduler and recent runs;
guides — built-in, self-documenting help (see Built-in guidance).
Connecting a client#
Any MCP client that speaks streamable HTTP can connect to the server’s
/mcp endpoint. There is nothing to install on the client side — point
your client at the endpoint URL as a remote HTTP MCP server.
Clients are configured either through a command that registers a remote HTTP server, or through a JSON configuration such as:
{
"mcpServers": {
"refinery": {
"type": "http",
"url": "https://your-refinery-host/mcp"
}
}
}
On first use the client performs an OAuth2 login in your browser against the refinery’s identity provider; subsequent requests carry the resulting token.
Authentication and roles#
Access uses the same OAuth2/OpenID Connect authentication as the rest of the
pro application, and the same four-tier role ladder — guest < ro <
rw < admin (see Securing the Refinery). Your role governs what the
assistant may do on your behalf:
read tools require the
ro,rworadminrole;write tools — anything that creates, modifies or deletes an object — require
rworadmin;guestcan call nothing.
In the tool catalog below, write tools are marked (write); all others are read-only.
Built-in guidance#
The server is designed to onboard an assistant on its own. The
get_*_guide and get_*_spec tools listed below return structured
instructions and language references, so a client can learn the query
language, the formula operators, or the figure/dashboard/balance syntaxes on
demand rather than from external documentation.
A single MCP prompt, welcome, acts as the entry point: it presents a
menu of activities — discover the instance, analyse the gold layer, detect
anomalies, create and test formulas, explore and manage dashboards — and
routes the assistant to the relevant guide. It also reminds the assistant to
check the notes at the start of a session, and to ask for explicit
confirmation before creating or changing anything.
Available tools#
Notes — a knowledge base the server keeps for the instance.
list_notes— list all note keys.get_note— read one note.save_note(write) — create or update a note (full replace, append, or find-and-replace).delete_note(write) — remove a note.
Note
The notes are a collaborative, instance-wide knowledge base: everyone using the instance shares the same notes. An assistant can record what it learns about the refinery — naming conventions, a domain map, recurring gotchas — and that knowledge persists across sessions and is available to every other user.
Discovery, search
find_items— search series or groups with the lisp query language.get_info— global statistics about the instance.list_metadata_keys— all metadata keys in use.get_dependents— what depends on an item (impact analysis).troubleshoot_item— metadata, internal metadata and, for formulas, depth and dependencies.
Series
get_fresh_data_sample— a recent data window with statistics (auto-downsampled to a readable size).analyze_series— change-point segmentation into regimes (flat, ascending, oscillating, …).get_activity_summary— per-series statistics across a basket, ranked by variability.rename_series(write) — rename a series or formula, propagating into referencing formulas.delete_series(write) — delete a series or formula (irreversible).update_metadata(write) — merge metadata onto a series or group.
Formulas
get_formula— a formula definition, optionally expanded.eval_formula— evaluate an ad-hoc formula without saving it.get_gold_layer_formulas— the deepest, highest-value formulas (expensive).register_formula(write) — create or replace a series or group formula.
Baskets
list_baskets— saved queries and their definitions.resolve_basket— evaluate a basket to its matching items.register_basket(write) — create or update a basket.delete_basket(write) — remove a basket.
Figures
list_figures— the figure catalog (paginated, by variant).figure_dependents— dashboards referencing a figure.get_figure— a figure’s canonical definition.update_figure(write) — create or replace a figure (validated before saving).rename_figure(write) — rename, rewriting referring dashboards.delete_figure(write) — delete a figure (refuses if referenced, unless forced).
Dashboards
list_dashboards— dashboards with title and description.get_dashboard— a dashboard’s canonical definition.update_dashboard(write) — create or replace a dashboard.rename_dashboard(write) — rename a dashboard.delete_dashboard(write) — delete a dashboard.
Balances
list_balances— balance definitions in the registry.get_balance— a balance’s canonical definition.update_balance(write) — create or replace a balance (validated).rename_balance(write) — rename, rewriting referencing figures.delete_balance(write) — delete a balance (refuses if referenced).
Tasks and schedules
list_operations— registered scheduler operations.list_schedules— scheduled rules and their decoded inputs.list_tasks— recent task runs with their status.
Guides and help
get_query_syntax— the search query language.get_operators— the formula operators (introspected from the registry).get_discovery_guide— a methodology to map an unfamiliar refinery.get_formula_building_guide— a step-by-step formula authoring workflow.get_gold_layer_analysis_guide— how to find and read the highest-value formulas.get_anomaly_detection_guide— a workflow for spotting outliers and gaps.get_figure_spec— the figure definition language.get_dashboard_spec— the dashboard definition language.get_balance_spec— the balance definition language.