# Migrate Meltano Connector > Migrate an existing Meltano extractor or Singer tap workflow to a Fivetran Connector SDK connector. Use when the user has a Meltano project, meltano.yml extractor, custom Singer tap, catalog/state/config files, or tap source code they want to port to CSDK. - Skill: `fivetran/migrate-meltano-connector` (Agent Skill) - Install (CLI): `npx skillmds@latest add fivetran/migrate-meltano-connector` - Raw SKILL.md: https://api.skillmd.com/api/skills/fivetran/migrate-meltano-connector/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - Author: fivetran (https://skillmd.com/u/fivetran) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/fivetran/migrate-meltano-connector --- > **Context**: This plugin is for the Fivetran Connector SDK (CSDK). "CSDK" is shorthand for "Connector SDK". # Migrate a Meltano Extractor or Singer Tap to CSDK **FIRST**: Read `sdk-reference.md` from the plugin directory to load SDK rules, operation names, schema constraints, configuration rules, and testing requirements. Use this workflow to convert Meltano extractor logic into a Connector SDK project. Meltano migration references: - Meltano project/config docs: https://docs.meltano.com/concepts/project/ - Meltano plugin docs: https://docs.meltano.com/concepts/plugins/ - Singer spec reference: https://hub.meltano.com/singer/spec/ - Meltano Singer SDK docs: https://sdk.meltano.com/ - Fivetran CLI for pipeline resources: https://pypi.org/project/fivetran-cli/ When validating the migration workflow itself, use representative public projects such as `Matatika/example-github-analytics`, which exercises: - A Meltano project with `tap-github`, `target-postgres`, dbt transforms, a utility plugin, a job, and a schedule. - Selector rules like `*!.*` followed by explicitly selected streams. - Array settings such as `repositories`. - A third-party Singer SDK tap whose source code must be consulted for schemas, primary keys, parent-child stream context, pagination, and state behavior. ## Step 1: Locate the Source and Target If the user did not provide a source path, ask for one of: - A Meltano project directory containing `meltano.yml`. - A custom Singer tap package or local tap source directory. - Singer artifacts: `config.json`, `catalog.json` or `properties.json`, `state.json`, and sample output. Find or create the target CSDK project: - If the current directory already has `connector.py`, `configuration.json`, and `requirements.txt`, migrate in place. - If no CSDK project exists, ask for the target directory name, then scaffold it with `fivetran init` before editing files. - Do not overwrite unrelated files. Scope the migration: - Migrate Meltano extractors/Singer taps only. - Do not port Meltano loaders/targets, dbt transforms, schedules, orchestration, or environments into CSDK connector code. - For pipeline resources that correspond to Fivetran platform objects, create a follow-up plan using `fivetran-cli` instead of embedding that behavior in `connector.py`. - Use loaders, transforms, schedules, and environments as context for what the extractor produced, how it was configured, and which Fivetran CLI resources may need to be created or updated. ## Step 2: Inventory the Meltano Source Read source files and identify: - Meltano project files: `meltano.yml`, lock files under `plugins/`, `.env` templates, catalog files, state files, and any local tap source. - Extractor name, namespace, executable, variant, `pip_url`, capabilities, and settings. - Whether the tap is a local/custom tap, a third-party package, or a MeltanoHub discoverable extractor. - Selected streams and fields from `select`, catalog metadata, or catalog `selected` flags. - Selection rule semantics and precedence. For example, `*!.*` followed by `issues.*` means start with all streams deselected, then include the named streams. - Stream schema from discovery/catalog JSON, tap source classes, or sample Singer `SCHEMA` messages. - Primary keys from `key_properties`, `table-key-properties`, SDK stream `primary_keys`, or source code. - Replication method and replication keys from metadata, SDK stream settings, or tap source code. - State/bookmark shape from `state.json`, Singer `STATE` messages, or tap source code. - Tap configuration fields, defaults, env aliases, and which fields are credentials. - Complex settings such as arrays and objects. Prefer separate connector deployments for multi-entity settings; record JSON-encoded string handling only when it is needed for parity with the source tap. - API/database/file access logic, pagination, rate limiting, retries, and error handling. - Parent-child stream dependencies. Some selected child streams require parent stream API calls for context even when the parent stream is not selected for output. - Pipeline intent from loaders, targets, transformations, schedules, and environments. Record this separately from extractor migration as potential `fivetran-cli` follow-up work. Do not ask the user to paste credentials in chat. Do not print `.env`, `config.json`, or Meltano config values. Replace real values with placeholders if they appear in source files. ## Step 3: Map Meltano/Singer Concepts to CSDK Use this mapping: | Meltano/Singer concept | Connector SDK concept | |------------------------|-----------------------| | Meltano extractor / Singer tap | One CSDK connector, or one scoped connector per source if the project has multiple unrelated extractors | | `meltano.yml` extractor settings / tap `config.json` | Flat string fields in `configuration.json` | | Meltano array/object settings | Prefer separate connector deployments for multi-entity sync; use JSON-encoded string fields parsed with `json.loads()` only when unavoidable for source-tap parity | | Meltano environment variables and setting aliases | Documented configuration fields; do not embed environment loading into connector code | | Singer catalog stream | CSDK `schema(configuration)` table entry | | Singer `key_properties`, `table-key-properties`, or stream `primary_keys` | CSDK `primary_key` | | Singer JSON Schema properties | Optional CSDK `columns` with SDK data types | | Selected streams/fields | Tables and fields implemented by `schema()` and `update()` | | Parent streams used only for context | Internal helper/API calls; do not emit the parent table unless it was selected | | Singer `RECORD` message | `op.upsert(table=stream, data=record)` | | Singer `STATE.value` / bookmarks | `op.checkpoint(state=new_state)` | | Singer incremental replication key | Cursor logic inside `update(configuration, state)` | | Singer discovery mode | Static or configuration-aware `schema(configuration)` | Map JSON Schema types conservatively: | JSON Schema | CSDK column type | |-------------|------------------| | `boolean` | `BOOLEAN` | | `integer` | `LONG` | | `number` | `DOUBLE` | | `string` with `format: date-time` | `UTC_DATETIME` | | `string` with `format: date` | `NAIVE_DATE` | | `string` | `STRING` | | `object` or `array` | `JSON` | | multiple types including `null` | Use the non-null type | | unknown or mixed type | Omit the column type or use `JSON` when the field is structured | Full-table behavior needs explicit review. Singer taps commonly emit current rows without an explicit delete stream. If the old Meltano pipeline relied on a target to replace/truncate data for full-table streams, document that behavior and only use `op.truncate(table=...)` when the intended CSDK behavior is to mark missing prior rows as deleted before reloading a complete snapshot. ## Step 4: Port the Implementation Edit the CSDK project files: ### `connector.py` - Keep the scaffolded structure where present: `validate_configuration()`, `schema()`, `update()`, global `connector = Connector(...)`, and `if __name__ == "__main__": connector.debug()`. - Port extraction logic directly into Python helper functions. - If the source tap is already Python, reuse small pure extraction helpers where practical, but remove Singer CLI, stdout message emission, and Meltano runtime assumptions. - If only a Meltano plugin declaration exists and no tap source is available, treat it as a behavior/configuration specification and reimplement from the source API/database/file protocol. - Replace tap config access with `configuration`. - Replace Singer state/bookmarks with the CSDK `state` dict. Preserve the original bookmark shape when it makes parity easier. - Emit rows using SDK operations directly. Do not `yield` operations and do not print Singer JSON messages. - Call `op.checkpoint(state=state)` after each safe page, batch, or stream boundary. - Preserve selected stream behavior; do not silently add streams that were disabled in Meltano. - Preserve incremental cursor semantics, including start-date behavior for empty state. - Add retry handling for HTTP 429 and transient 5xx responses. - Keep type hints simple: use `dict` and `list`; do not import `Dict`, `Any`, or use `op.Operation`. ### `configuration.json` - Keep flat string key/value pairs only. - Include fields needed by the connector; preserve supplied values and use obvious placeholders for unresolved fields. - Convert nested Meltano settings to clear flat names, and document any rename. - Prefer separate connector deployments for multi-entity configs (see `sdk-reference.md`). Only when the source tap truly requires multi-item selection, represent Meltano arrays or objects as JSON-encoded string placeholders, such as `"repositories": "[\"owner/repo\"]"` or `"searches": "[{\"name\":\"example\",\"query\":\"repo:owner/repo\"}]"`, then parse and validate them with `json.loads()` in `connector.py`. - Do not include real credentials from `.env`, `config.json`, or Meltano config. - Do not use arrays or nested objects. ### `requirements.txt` - Include only source-specific dependencies not already available in the CSDK environment. - Do not include Meltano, Singer SDK, Singer target packages, or `fivetran_connector_sdk`. - Do not include loader/target/dbt/orchestrator dependencies. ### `README.md` - Explain that this connector was migrated from a Meltano extractor/Singer tap. - Document source extractor name, selected streams, primary keys, and cursor fields. - Document configuration fields with placeholders only. - Document any behavior changes, especially full-table replacement/truncate semantics, stream selection changes, and removed Meltano loader/transform/orchestration behavior. - Document any pipeline resources that were intentionally left out of connector code and should be handled with `fivetran-cli`, such as destination/connection setup, transformations, or transformation projects. - Follow **Configuration entry** in `sdk-reference.md`; use the SDK form or ordinary JSON values. ### `fivetran-cli` follow-up notes - If the Meltano pipeline includes loader/target configuration that maps to Fivetran destinations or connections, document the intended Fivetran CLI resource commands to run after connector migration. - If the Meltano pipeline includes dbt transformations or transformation projects, document the intended Fivetran CLI transformation or transformation-project commands to run after connector migration. - Do not execute account-changing `fivetran-cli` commands unless the user explicitly asks for deployment/setup and provides the required non-secret identifiers. Never ask the user to paste API keys in chat. ## Step 5: Validate the Migration Check behavior before testing: - Every selected stream has a CSDK schema entry with a primary key. - Disabled streams are not emitted as tables. If a disabled parent stream is needed for child context, it is used only as an internal helper path. - Every Singer `RECORD` stream maps to `op.upsert`. - Every incremental stream checkpoints after a safe cursor boundary. - Array/object configuration fields are documented as JSON-encoded strings and parsed safely. - Full-table streams have an explicit documented decision: upsert-only snapshot or `op.truncate` plus reload. - JSON Schema to CSDK column type mapping is conservative and does not invent unsupported SDK types. - No Singer stdout message generation remains. - No Meltano loader, target, dbt, schedule, or environment code is required for local CSDK execution. - Any Meltano pipeline resources that should survive migration are listed as `fivetran-cli` follow-up work instead of being embedded in connector code. - No real credentials are present in source files, README, or chat. Then follow the secure test flow from `test-connector`: - Run the secure runner, not `fivetran debug` directly. - If configuration values are missing, follow **Configuration entry** in `sdk-reference.md`; preserve supplied and existing values. - Do not inspect or print configuration values. ## Step 6: Report Results Summarize: - Source Meltano project or tap package migrated. - Streams migrated, primary keys, and cursor fields. - Configuration fields created or renamed. - State/bookmark mapping. - Full-table replacement/truncate decision. - Meltano pieces intentionally not migrated into connector code, such as loaders, transforms, schedules, and environments. - `fivetran-cli` follow-up plan for pipeline resources, if applicable. - Remaining manual checks, especially API credentials, endpoint access, and data parity against the original Meltano run.