Conductor
Conductor is a cloud development platform for creating workspaces, running coding-agent sessions, exchanging messages, and searching session transcripts.
Verdict
Common use cases
- Audit active agent sessions before month-end
- Archive idle workspaces to cut cloud spend
- Retrieve session transcripts for compliance review
- Spin up isolated environments for new projects
- Check workspace initialization status during deploys
Integration
- Vendor
- Conductor
- Category
- other
- Auth
- API_KEY
- Tools
- 17
- Composio slug
conductor
Tools
- Change Workspace State
Put a workspace to sleep or archive it. Both actions stop active workspace use; archive is a stronger transition but is documented as restorable. Confirm the intended target and state before calling.
- Create Session
Create another agent chat session in an existing workspace. This may start agent/model resources and incur charges; call only when the user explicitly intends a new session.
- Create Workspace
Create a cloud workspace and its first agent session. This launches compute and may incur cloud or model charges; call only when the user explicitly intends to create a new workspace.
- Get Current User
Return the user and organization represented by the connected Conductor API key. Use this to verify which account the connection acts as.
- Get Message
Get one transcript message by id. The content may contain sensitive prompts, agent output, configuration, or provider-specific events.
- Get Project
Get one Conductor project and its repository remote by project id.
- Get Session
Get one agent session's configuration and current idle, working, or error status in one read. The tool fails unless both records are retrieved and refer to the requested session.
- Get Workspace
Get a workspace's metadata and current lifecycle status in one read, including initialization progress or errors. The tool fails unless both records are retrieved and refer to the requested workspace.
- List Messages
Read one page of a session transcript or incrementally read messages after a known message ID. Returned content may contain sensitive user prompts, agent output, configuration, or provider events; request and expose it only when needed.
- List Projects
List repositories available to the connected account for creating Conductor workspaces.
- List Sessions
List one page of agent chat sessions in a Conductor workspace.
- List Workspaces
List one page of workspaces belonging to a Conductor project, including deep links and recent activity.
- Query Transcripts
Run one read-only SQL SELECT against the organization's session_transcripts_view. Results can expose sensitive transcript text across many sessions: select only necessary columns, constrain rows with LIMIT, and use this only for deliberate
- Rename Session
Change an existing Conductor session's display name without stopping its agent.
- Rename Workspace
Change the display name of an existing Conductor workspace without changing its lifecycle state.
- Send Message
Send a prompt to a Conductor session agent. This can start external agent/model execution and incur charges; it is not a passive transcript append. Call only when the user intends the agent to act.
- Stop Session
Cancel only the current turn and queued messages, or archive the entire session. Both can discard queued work; archive also stops the agent and leaves the session read-only. Confirm the intended action before calling.
Setup
Setup guide
- 11. In Switchy, open your workspace settings and navigate to the Integrations tab. 2. Click 'Add MCP Integration' and select Conductor from the list. 3. You'll be prompted to paste a Conductor API key — generate one from your Conductor dashboard under API settings. 4. Grant the key read and write access to workspaces, sessions, and projects. 5. Click 'Connect' and wait for Switchy to verify the key. 6. Open any Project in Switchy and type '@Conductor get current user' to confirm the connection returns your organization details. 7. To invoke a tool, @mention Conductor in a message and describe the action — for example '@Conductor create a new workspace for the Q2 demo' or '@Conductor show me all active sessions'. 8. Review the tool confirmation before Switchy executes, especially for create or archive actions that affect billing.
What teammates see: by default, memories from Conductor are scoped to the Project (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Projects have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
List Active Sessions
@Conductor show me every active session in our organization and their current status
Create Demo Workspace
@Conductor create a new workspace called 'Client Demo March' and start the first session
Archive Idle Workspaces
@Conductor find workspaces that have been idle for more than 7 days and archive them
Retrieve Session Transcript
@Conductor get the complete transcript for session ID abc123 including all prompts and outputs
Check Workspace Status
@Conductor get the current status and initialization progress for workspace ID xyz789
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@conductor show me all workspaces in my account and their current status
Your Conductor account has 4 active workspaces: 1. **prod-api-refactor** (active) – Session running Claude 3.5 Sonnet, last activity 12 minutes ago 2. **docs-migration** (sleeping) – Last used 3 days ago, can be woken on demand 3. **prototype-ui** (active) – Session running GPT-4, currently idle 4. **legacy-cleanup** (archived) – Archived 2 weeks ago, restorable if needed All workspaces are linked to the 'acme-eng' organization. The two active workspaces are incurring compute charges.
This example uses Get Workspace and Get Current User to audit your Conductor footprint. Useful before creating new workspaces or when tracking cloud spend. Requires an API key with read access to all workspaces in the organization.
@conductor put the docs-migration workspace to sleep until we need it again next quarter
Workspace **docs-migration** has been put to sleep. The agent session is stopped and compute resources are released. No further charges will accrue until you wake the workspace. You can restore it anytime with a wake command. The workspace's project files, session history, and configuration remain intact. Archived workspaces require a stronger restore action; sleep is the reversible pause you wanted here.
This example calls Change Workspace State to stop incurring charges without losing work. Sleep is reversible; archive is stronger but also restorable. Confirm the workspace name before running—this action immediately stops the session and may interrupt in-progress agent work.
@conductor create a new workspace called 'q1-planning' with a Claude session, then draft a project kickoff outline based on our last three sprint retros
Workspace **q1-planning** is created and a Claude 3.5 Sonnet session is starting. Initialization will complete in ~30 seconds. Once the session is live, I'll retrieve your recent sprint retro transcripts from the other workspaces and draft a kickoff outline covering: team velocity trends, recurring blockers, and proposed Q1 goals. This will be a new session, so it won't have context from your other workspaces unless we explicitly load it. Note: This workspace will incur compute and model API charges from creation until you sleep or archive it.
This example chains Create Workspace with a synthesis task that spans multiple sessions. Conductor workspaces are isolated by default, so cross-workspace context requires explicit retrieval. Be aware that workspace creation starts billing immediately—confirm intent before calling this tool.
Use-case deep-dives
When you need parallel test agents without local setup
A 6-person QA team runs regression tests against three LLM configurations every sprint. Conductor wins here because Create Workspace and Create Session let you spin up isolated agent environments on demand, then archive them when the test cycle ends. Each workspace maps to one test scenario; each session is one model variant. The team avoids local Docker sprawl and the API key auth keeps credentials out of CI logs. Trade-off: if your test matrix exceeds 20 concurrent workspaces, you'll hit rate limits and need to batch. If you're testing a single agent config or running tests serially, a local container is simpler. This MCP pays off when you need elastic, short-lived agent environments that multiple people trigger from Slack or a dashboard.
Fast throwaway workspaces for sales engineering
A 3-person sales-eng team demos a conversational AI product to enterprise prospects. Conductor is the right call when you need to create a clean workspace per demo call, pre-load a project repo, and tear it down after. Get Current User confirms which org the API key belongs to before you create resources; Get Workspace and Get Session let you verify the demo is live before screen-sharing. Change Workspace State archives the environment post-call so you're not paying for idle compute. The 17-tool surface is overkill if you only need read access to transcripts—Get Message is useful for debugging but not core to the demo flow. This setup works when demos are frequent, short, and need isolated state. If you demo the same persistent environment to all prospects, skip the MCP and use a shared staging instance.
When you need to pull session logs into a SIEM
A 2-person compliance team at a healthcare startup audits agent conversations for PHI leakage. Conductor's Get Message and Get Session tools let you export transcript content and session metadata into your log aggregator without SSH-ing into production. Get Project ties each session back to the code version that ran it, which matters for change-control audits. The API key auth is fine for batch jobs but risky if you're calling these tools from a shared Slack channel—treat the key like a database credential. Limitation: if your session count is over 10k per month, paginating through Get Session calls gets slow; you'll want the vendor's bulk-export API instead. This MCP is the move when you need programmatic access to a few dozen sessions per week and your compliance tooling can ingest JSON.
Frequently asked
What does the Conductor MCP let me do in Switchy?
It lets your team create and manage Conductor cloud workspaces, start agent sessions, and retrieve session transcripts — all from Switchy chat. You can put workspaces to sleep or archive them, check which user the API key belongs to, and pull project metadata. Think of it as remote-controlling your Conductor infrastructure without leaving the conversation.
Do I need admin access to connect Conductor?
You need a Conductor API key with permission to create workspaces, manage sessions, and read transcripts. Conductor uses API_KEY auth, so whoever generates the key controls what the MCP can do. If your org restricts key creation to admins, you'll need one of them to provision it for Switchy.
Can the MCP read my agent session transcripts?
Yes. The Get Message tool retrieves individual transcript messages by ID, including prompts, agent output, configuration, and provider events. If your sessions contain sensitive data, treat the API key like you would direct database access — only connect it in Switchy workspaces where that visibility is acceptable.
Why use this instead of the Conductor web UI?
The MCP is faster for repetitive ops — spinning up a workspace, checking session status, archiving old environments — especially when you're already in a Switchy chat coordinating with your team. The web UI is still better for deep config or browsing all workspaces; the MCP is for inline automation and status checks.
Who on the team should connect the Conductor MCP?
Whoever provisions cloud resources or manages agent infrastructure. The MCP can create workspaces and sessions, which launch compute and incur charges, so you want someone who understands your Conductor billing and lifecycle policies. Don't hand the key to people who just need read-only session logs.