🐛 Bug: Coral MCP tries to install connectors before deploying a coral cloud destination

[aaronsteers]

Coral MCP tries to install connectors before deploying a Cloud destination. This can fail if the connector can't be installed, for instance if the Python version is a mismatch with the connector, or if PyAirbyte doesn't have a writeable directory.

Workarounds

First workaround is to ensure that the AIRBYTE_PROJECT_DIR env var is set as described here: https://docs.airbyte.com/ai-agents/pyairbyte-mcp#step-1-generate-a-dotenv-secrets-file

Second workaround is to use a CLI arg like --python=3.11 to ensure a compatible Python version is being used. Example here: https://docs.airbyte.com/ai-agents/pyairbyte-mcp#step-2-registering-the-mcp-server

Third workaround is to try to force Docker mode instead of Python. (Not well documented.)

Proposed Fixes:

The best fix is to skip installation when we are only creating the source or destination in order to deploy it to Cloud. (This might be implemented already for sources, and perhaps just destinations are not patched to leverage this pattern?)

Another fix option is to force Docker mode when we are running as MCP. I had thought this was already implemented, but the setting may not be correctly applied across both sources and destination use cases.

Docs: https://airbytehq.github.io/PyAirbyte/airbyte/mcp.html

[coderabbitai]

A summary of the changes CodeRabbit can apply:

• Add a "Deploying to Airbyte Cloud" section to airbyte/destinations/init.py that documents MCP's use of install_if_missing=False (so connectors are not installed locally) and clarifies that AIRBYTE_PROJECT_DIR, specific Python versions, or local Docker are not required for cloud deployments.

• Add a "Deploying to Airbyte Cloud" documentation section to airbyte/destinations/init.py explaining automatic cloud deployment via MCP, showing a code example that uses get_destination(..., install_if_missing=False) and workspace.deploy_destination(...), and clarifying that setting install_if_missing=False prevents local calls like check() or write() but allows deploying the destination configuration to Airbyte Cloud.

• ✅ Create PR with these edits
• 📋 Get copyable edits