Tool reference

The server registers 14 tools. In your host's tools/list they appear as a flat list; here they're grouped by journey for readability. Each tool declares MCP annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) so your host knows when to ask for confirmation.

Build

build_system

Build a brand-new system from a description. Returns a preview; it does not save. You must call confirm_save(build_id) after approving.

• Scope: workspace:build
• Annotations: not read-only, not destructive, open-world, not idempotent (send an Idempotency-Key)
• Streams: yes (progress as it plans and builds)

Input

{
  "intent": "a lead capture form, a table to store entries, and a workflow that pings me on Slack",
  "preferred_integrations": ["slack"],
  "extend_build_id": null
}

Output

{
  "build_id": "bld_...",
  "status": "preview_ready",
  "summary": "Form + table + Slack follow-up workflow",
  "assets": [{ "type": "form", "name": "Lead Capture", "...": "..." }],
  "warnings": [],
  "requires_oauth": [],
  "next_action": { "tool": "confirm_save", "arguments": { "build_id": "bld_..." } },
  "expires_at": "2026-05-30T12:00:00Z"
}

Example: "Build me a lead intake system: a form, a table to store entries, and a Slack follow-up." → build_system(intent) → preview with a build_id → you approve → confirm_save(build_id).

update_asset

Modify an existing asset. Like build_system, it returns a preview to approve with confirm_save.

• Scope: workspace:build + workspace:read
• Annotations: not destructive, open-world, not idempotent
• Streams: yes

Input

{ "asset_id": "ast_...", "intent": "add an auto-responder email when someone submits" }

Output: same shape as build_system; the summary describes the diff.

confirm_save

Persist a previewed build. Only call after the user explicitly approves.

• Scope: workspace:write
• Annotations: destructive, idempotent (calling twice with the same build_id returns the same result)

Input

{ "build_id": "bld_...", "publish_immediately": false }

Output

{
  "status": "saved",
  "assets": [{ "asset_id": "ast_...", "type": "form", "product_url": "https://..." }],
  "push_errors": [],
  "message": "Saved 3 assets"
}

status may be saved, partial_save, expired, or already_saved. Previews expire one hour after the build.

abort_build

Cancel an in-flight build_system or update_asset. Per the MCP spec, a dropped connection is not a cancellation; use this tool to intentionally stop a build.

• Scope: same as the original build
• Annotations: destructive, idempotent

Input: { "build_id": "bld_..." } Output: { "status": "aborted" | "already_complete" | "not_found", "message": "..." }

Discover

list_my_assets

List assets in the workspace. Use it before suggesting new builds, or to find an asset_id for update_asset and the operate tools.

• Scope: workspace:read
• Annotations: read-only, idempotent

Input

{ "type": "all", "limit": 50, "cursor": null }

type is one of form, workflow, table, email, agent, sequence, or all.

Output

{
  "assets": [
    { "asset_id": "ast_...", "type": "workflow", "name": "Onboarding", "publish_status": "published", "product_url": "https://..." }
  ],
  "next_cursor": null
}

describe_my_workspace

Workspace context: connected integrations, asset counts, recent activity, and credit balance. Call this before build_system when the request is open-ended.

• Scope: workspace:read
• Annotations: read-only, idempotent

Input: {} Output

{
  "workspace": { "id": "ws_...", "name": "Acme" },
  "asset_counts": { "form": 4, "workflow": 7, "table": 3 },
  "connected_integrations": ["gmail", "slack"],
  "credit_balance": 1240,
  "recent_builds": []
}

Operate

query_records

Query records from a table with filters, sorts, field projection, and pagination.

• Scope: workspace:read
• Annotations: read-only, idempotent

Input

{
  "table_id": "tbl_...",
  "filters": [{ "field": "status", "op": "eq", "value": "active" }],
  "sort": [{ "field": "created_at", "dir": "desc" }],
  "fields": ["name", "company", "status"],
  "limit": 50
}

Filter operators: eq, neq, gt, lt, gte, lte, contains, in, is_empty, is_not_empty.

Output

{ "records": [{ "id": "rec_...", "fields": { "name": "Ada", "status": "active" } }], "total_estimate": 12, "next_cursor": null }

add_record

Add a single record to a table.

• Scope: workspace:execute
• Annotations: not destructive, not idempotent; send an Idempotency-Key (calling twice otherwise creates two rows)

Input: { "table_id": "tbl_...", "fields": { "name": "Ada Lovelace", "company": "Analytical" } } Output: { "record_id": "rec_...", "fields": { "...": "..." }, "created_at": "..." }

update_record

Update fields on an existing record.

• Scope: workspace:execute
• Annotations: destructive, idempotent

Input: { "table_id": "tbl_...", "record_id": "rec_...", "fields": { "status": "contacted" } } Output: { "record_id": "rec_...", "fields": { "...": "..." }, "updated_at": "..." }

trigger_workflow

Trigger a published workflow with a payload. Runs synchronously (workflows typically finish in under 30 seconds).

• Scope: workspace:execute
• Annotations: destructive, open-world, not idempotent

Input

{ "workflow_id": "ast_...", "payload": { "email": "ada@analytical.com", "plan": "pro" } }

Output

{ "execution_id": "exe_...", "status": "completed", "result": { "...": "..." }, "duration_ms": 4200 }

Common errors: workflow_not_found, workflow_not_published, workflow_failed, execution_timeout. See Errors.

send_via_integration

Invoke one operation on a connection the user has already set up (Gmail, Slack, HubSpot, …). For one-shot sends and posts, not for building workflows.

• Scope: workspace:execute
• Annotations: destructive, open-world, not idempotent (send the same email twice and you spam)

Input

{
  "integration": "gmail",
  "operation": "send-email",
  "params": { "to": "ada@analytical.com", "subject": "Following up", "body": "Hi Ada…" },
  "connection_id": null
}

Output: { "status": "succeeded", "integration_response": { "...": "..." }, "used_connection_id": "conn_..." }

Common errors: integration_not_connected (response includes a connect_url), operation_not_found, oauth_token_expired, provider_error.

run_agent

Send a message to a published agent and stream the reply.

• Scope: workspace:execute
• Annotations: not destructive, open-world
• Streams: yes

Input: { "agent_id": "agt_...", "message": "Draft a reply to this ticket: …", "thread_id": null } Output: { "thread_id": "thr_...", "response": "…", "tool_calls_summary": [] }

To continue a conversation, pass the returned thread_id on the next call.

run_enrichment

Submit a person, company, or email for enrichment. Returns a job handle immediately; poll with check_enrichment.

• Scope: workspace:execute
• Annotations: not destructive, idempotent
• Async: yes

Input: { "type": "email", "input": { "email": "ada@analytical.com" } } Output: { "job_id": "job_...", "status": "submitted", "estimated_completion_seconds": 30 }

check_enrichment

Check the status of a previously submitted enrichment job.

• Scope: workspace:read
• Annotations: read-only, idempotent

Input: { "job_id": "job_..." } Output

{ "job_id": "job_...", "status": "completed", "result": { "company": "Analytical Engines Ltd", "...": "..." }, "completed_at": "..." }

status is pending, completed, or failed.