MCP: return structured results from Cloud deploy/connection/sync tools instead of prose strings

[devin-ai-integration]

Problem

Several Cloud MCP tools that create resources or start jobs return a prose sentence (-> str) instead of a structured object. An agent has to parse IDs and URLs out of free text, for example:

Sync started. Job ID is '<job_id>' and job URL is: <job_url>

We noticed this while testing the local stdio MCP end to end against Airbyte OSS 2.3.0. The deploy → connect → sync flow worked, but each step's IDs had to be scraped from sentences before the next call could use them.

Affected tools (airbyte/mcp/cloud.py):

• deploy_connector_to_cloud (source and destination deploys)
• deploy_noop_destination_to_cloud
• create_connection_on_cloud
• run_cloud_sync

Proposal

Return Pydantic result models, as the read/list tools already do, so FastMCP emits structuredContent plus an output schema. Pseudocode:

class CloudSyncStartedResult(BaseModel):
    workspace_id: str
    connection_id: str
    job_id: int
    job_url: str
    status: str | None  # set when wait=True

class CloudConnectorDeployResult(BaseModel):
    workspace_id: str
    connector_id: str      # source_id / destination_id
    connector_name: str
    connector_url: str

This changes the tool contract (return type and schema), so it should be a deliberate change on release-candidate/v1. It is separate from the workspace fix in #1305.

Written by Devin