create-dataflow
ChatGPTCreate a new, empty dataflow — the container you attach a source and destination to. This is the first step of manual pipeline setup; it returns the dataflow id you pass to create-dataflow-source and create-dataflow-destination. If a template fits the user's request, prefer create-dataflow-from-template instead, which returns a pre-configured dataflow.
create-dataflow-destination
ChatGPTAdd a destination to an existing dataflow — where the data is written (e.g. Google Sheets, BigQuery). Prerequisites: a dataflow must already exist and the user must have a credential for the destination integration (check list-credentials). Take the integration key from list-integrations and its parameter values from get-integration / get-integration-field-options — do not guess them. Returns the created destination.
create-dataflow-from-template
ChatGPTCreate a new dataflow from an existing template. Use list-templates first to find available templates and their IDs. The created dataflow will have sources pre-configured from the template but some additional source and destination configuration may be needed before it is ready to run.
create-dataflow-source
ChatGPTAdd a source to an existing dataflow — the service data is pulled from (e.g. Stripe, PostgreSQL). Prerequisites: a dataflow must already exist (create-dataflow first) and the user must have a credential for the integration (check list-credentials). Take the integration key from list-integrations and its parameter values from get-integration / get-integration-field-options — do not guess them. Returns the created source.
create-dataset
ChatGPTCreate a SQL dataset for a specific dataflow. Use this when you need to define a new SQL view over existing datasets in the SAME dataflow. Provide the SQL statement in the query argument, referencing other datasets by their table_name (returned by list-datasets/search-datasets/create-dataset). Use get-schema on each dataset snapshot to discover column names; take each columnName verbatim and wrap it in double quotes, since names routinely contain spaces, dots and punctuation. The query is stored and re-runs on every refresh, so alias output columns with AS to pin the names this dataset exposes. If a stored query later fails with "no such column", an upstream column was relabelled — re-read get-schema instead of guessing a variant. The query must be read-only — a SELECT, or a WITH … SELECT. PIVOT must use the SQL-standard form; the simplified PIVOT … ON … USING syntax is not supported. ALWAYS include the link (preview_url) in your response as it is the only way for users to run the query and verify data correctness.
get-data
ChatGPTGet data from a Coupler.io dataset by running SQL against the data table (the same SQL used by the create-dataset tool); queries accept DuckDB SQL syntax. Pass the dataset's dataset_id and your SQL statement in the query argument. Run the get-schema tool first with the same dataset_id to understand the table's columns, types and data context before building your query. Use each column's columnName verbatim and wrap it in double quotes — names routinely contain spaces, dots and punctuation. When visualizing the data, do not try to read any files or fetch any URLs, just generate a static page and use the data you get from the tools.
get-dataflow
ChatGPTGet the current state of a Coupler.io dataflow by its ID, including its sources and destinations.
get-integration
ChatGPTGet detailed integration info including its parameters, auth requirements, and configuration options. Use type "source" or "destination" and the Coupler.io-specific integration key.
get-integration-field-options
ChatGPTResolve available options for an integration parameter. Use this when get-integration returns resolve_options_with_tool: true for a param. Provide the dependent param values listed in options_depends_on so the correct options are returned. When options_requires_credential is true, you must also provide credential_id.
get-schema
ChatGPTGet the column schema for a dataset — read it before querying with get-data. Returns each column's key, human-readable label, type, and columnName: the exact name to use in SQL. Take columnName verbatim — never assume a format — and always wrap it in double quotes, since names routinely contain spaces, dots and punctuation: SELECT "Campaign Name" FROM data. A column's name can change when it is relabelled, so read the schema again rather than reusing names from earlier. If ai_context is present, it explains what the dataset actually means — always read it and factor it into your response.
get-skill
ChatGPTGet a skill's full instructions by name. Call this after list-skills returns a matching skill. You must follow the returned procedure to complete the task.
list-credentials
ChatGPTList the user's existing credentials (authorized connections) for source or destination integrations. Use this as a prerequisite check before create-dataflow-source or create-dataflow-destination — those calls need an active credential for the integration and fail without one. If none exists for the integration the user wants, do not stop at "you have no credentials": the response includes a setup URL for connecting a new credential — share it inline in your reply and guide the user through the connect round-trip (see the get-started skill; fetch it with get-skill if not already loaded). Pass a provider to get a setup URL that links straight to that provider's connect page. Returns each credential with its provider, type, and status.
list-dataflows
ChatGPTDEPRECATED. Use "list-datasets" instead. Reconnect to refresh the list of available tools if needed.
list-datasets
ChatGPTList datasets across the user's dataflows. Returns up to 200 datasets plus the true total in meta.total_count — if total_count exceeds the number returned, you are seeing only the first 200; narrow with dataflow_id/filters, or use search-datasets when you know the name, source, or a keyword. Once you've picked a dataset, use get-schema to read its columns and get-data to query it. An empty result does not always mean a new workspace — a narrow dataflow_id/filter, or a search-datasets query, can return nothing too. The reliable new-user signal is no credentials (list-credentials empty); an empty unfiltered list-datasets is the backup. When both say the account is bare, don't report "no datasets" — fetch the get-started skill (get-skill) and guide the user's first setup.
list-integrations
ChatGPTList the source or destination integrations Coupler.io supports. Use this to find the integration key for the service the user wants, then pass that key to get-integration to learn its parameters and auth requirements before creating a source or destination. Pass type "source" or "destination" to scope the list.
list-skills
ChatGPTSearch Coupler.io skills for one that fits the user's request; following a skill gives better results than ad-hoc tool calls. Once you know what the user wants, call this before other Coupler.io tools, with query set to a few keywords such as the data source, domain or task (e.g. "stripe revenue", "google ads", "create dataflow"). The search covers skill names, descriptions, slugs, categories and data sources. Omit query only to browse every skill. Returns skill slugs and short descriptions. If one matches, even loosely, call get-skill with its slug and follow the procedure. Skip the search for pure platform questions and tiny direct reads.
list-templates
ChatGPTList templates available in Coupler.io. Use this to discover available templates, optionally filtered by a specific source, metric or category, provide the link to set up the template and provide insights about their contents.
list-workspaces
ChatGPTList the Coupler.io workspaces you can access. Use this to discover workspace ids, then pass workspace_id on workspace-scoped tool calls to choose which workspace to act on.
run-dataflow
ChatGPTRun a Coupler.io dataflow. Pick the run mode by what you need: - run_mode: "preview" samples each source (up to 500 rows) and resolves the dataset schemas, much faster than a full run. It writes nothing to destinations and does not count as a refresh. Use it while setting up: after you create or change a source, run a preview, then read get-schema for each dataset to confirm the columns match what the user asked for, before you build on them (for example a SQL dataset) or offer a full run. It is also required before get-schema when list-datasets shows NO_SCHEMA. You don't need to ask before a preview run. - run_mode: "full" (the default) refreshes all data from the sources and writes it to the destinations; it can take minutes. Call it only when the user has asked to run, refresh or import the dataflow, or has confirmed one after setup. Do NOT call it to "fix" stale data noticed during analysis — surface the freshness gap to the user and let them decide. A preview makes the schema available but no rows: get-data needs a completed full run. The tool returns only the dataflow id once the run is queued; read get-schema to see when the schema is ready. The dataflow_id parameter is optional — the system may resolve it automatically. Do not ask the user for a dataflow_id. If the dataflow cannot be resolved, the tool will return an error.
search-datasets
ChatGPTSearch for datasets by name, source, or general text query. Use this instead of list-datasets when looking for a specific dataset and you need to narrow down from a large list. Parameters can be combined — they filter with AND logic. Use get-data and get-schema on the results to inspect dataset contents.
update-dataflow
ChatGPTUpdate a dataflow's attributes including its name and its run schedule. Provide a name to rename the dataflow. Provide a schedule_config with an interval (and optional days, start time, active hours and timezone) to set the timing, or with enabled to pause/resume the existing schedule. Does not change the dataflow's sources or destinations, which have their own tools.
update-dataset
ChatGPTUpdate dataset metadata: the name and/or the AI context (markdown description). Use ai_context to persist what you learned about the dataset contents so it can be recalled in future sessions; pass null to clear it. Record conventions and meaning rather than ready-to-run queries: a column's name can change when it is relabelled, so any column name written here is a snapshot — check it against get-schema before relying on it in a later session.
update-dataset-query
ChatGPTUpdate the SQL query of a customSql dataset. Only datasets with type customSql support query updates. In your SQL query, reference other datasets from the SAME dataflow by their table_name (returned by list-datasets/search-datasets/create-dataset). Use get-schema on each dataset snapshot to discover column names; take each columnName verbatim and wrap it in double quotes, since names routinely contain spaces, dots and punctuation. The query is validated and dataset columns are re-detected. It is stored and re-runs on every refresh, so alias output columns with AS to pin the names this dataset exposes. If a stored query later fails with "no such column", an upstream column was relabelled — re-read get-schema instead of guessing a variant. The query must be read-only — a SELECT, or a WITH … SELECT. PIVOT must use the SQL-standard form; the simplified PIVOT … ON … USING syntax is not supported. ALWAYS include the link (preview_url) in your response as it is the only way for users to run the query and verify data correctness.
get-data
Claudeget-dataflow
Claudeget-schema
Claudelist-dataflows
Claude
