Tines Workflow Automation

Build and manage Tines stories (automated workflows) using the Tines MCP server tools.

Setup Check

This skill requires the Tines MCP server to be connected. Check if the tines MCP server is available.

If NOT connected, guide the user through setup — see mcp-reference.md for the full deployment guide. Quick summary:

1. Edit TENANT_URL in generate_mcp.py to match their tenant
2. Run python3 generate_mcp.py to create story_export.json
3. Import into Tines:

   curl -X POST -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
     "$TENANT_URL/api/v1/stories/import" \
     -d '{"new_name":"claude-mcp","team_id":TEAM_ID,"data":'$(cat story_export.json)'}'

4. Create a Tines Credential named tines_api_key with the API key value, and add the tenant domain to allowed_hosts
5. Add the MCP server to Claude Code:

   claude mcp add -t http -s project tines \
     -H "Authorization: Bearer MCP_SECRET" \
     "TENANT_URL/mcp/MCP_PATH"

Core Concepts

• Story = A workflow (container for actions)
• Action = A step in the workflow (API type uses Agents:: prefix)
• Link = Connection between actions (data flows source -> receiver)
• Event = Data payload that flows through actions
• Resource = Shared data store (JSON/text/file) accessible across stories
• Credential = Stored authentication (API keys, OAuth tokens, etc.)
• Value Pills = <<action_slug.field.path>> syntax for referencing upstream data
• Formulas = Built-in functions like DATE(), IF(), SHA256(), etc.

MCP Tools Reference

Story Management

Action Management

Resources (Data Stores)

Story Notes

Story Events

Discovery

Workflow Testing

Parameter Conventions

All MCP tool parameters are strings. For complex values:

• options_json — Pass action options as a JSON string: "{\"mode\":\"message_only\",\"payload\":{\"key\":\"value\"}}"
• source_ids_json — Pass as JSON array string: "[\"12345\"]" (DEFAULT links only)
• links_to_sources_json / links_to_receivers_json — Pass as JSON array of objects: "[{\"source_id\":\"12345\",\"type\":\"NO_MATCH\"}]" (supports typed links)
• tags_json — Pass as JSON array string: "[\"tag1\",\"tag2\"]"
• value_json — Pass resource values as JSON string
• data_json — Pass full story export as JSON string (for import_story)

Formula escaping: When options_json values contain Tines formula references (e.g., <<webhook.body.field>>), escape the angle brackets as JSON unicode: use \u003c\u003c instead of << and \u003e\u003e instead of >>. This prevents premature formula evaluation during action creation. The Tines API stores these as the correct <</>> characters.

Action Types

When using create_action, the type parameter must be one of:

See action-types.md for detailed options_json schemas for each type.

Linking Actions

When creating actions, pass source_ids_json to link to upstream actions (creates DEFAULT links):

create_action(story_id="123", type="Agents::EventTransformationAgent",
              name="Process", options_json="{...}",
              position_y="150", source_ids_json="[\"456\"]")

Typed links (NO_MATCH, FAILURE) use links_to_sources_json or links_to_receivers_json instead:

create_action(story_id="123", type="Agents::EventTransformationAgent",
              name="Handle No Match", options_json="{...}",
              position_y="300", position_x="300",
              links_to_sources_json="[{\"source_id\":\"<trigger_id>\",\"type\":\"NO_MATCH\"}]")

Link types: DEFAULT, NO_MATCH (TriggerAgent only), FAILURE (action errored).

source_ids_json and links_to_sources_json are mutually exclusive (same for receiver_ids_json / links_to_receivers_json).

To remove links, use the delete_links tool:

delete_links(links_json="[{\"source_id\":\"456\",\"receiver_id\":\"789\"}]")

Layout Positioning

Use position_x and position_y when creating actions:

• Vertical spacing: ~150px between sequential actions
• Horizontal spacing: ~300px for parallel branches
• Start at (0, 0) for the first action

Value Pills & Formulas

Reference upstream data: <<action_slug.field.nested_field>> Array access: <<action_slug.items[0].name>>

Formulas

Common: IF(), DEFAULT(), SIZE(), JSON_PARSE(), TO_JSON(), DATE(), REGEX_EXTRACT(), MAP(), WHERE(), MERGE()

See formulas.md for the full formula reference (logic, text, arrays, objects, dates, crypto, numbers).

Common Mistakes

Building a Story — Process

1. Find your team: list_teams to get the team_id
2. Create or find the story: create_story or list_stories + get_story
3. Create actions top-to-bottom: Use create_action for each, linking with source_ids_json
4. Verify: export_story to check the full structure
5. Test: run_action on the first executable action (Event Transform, HTTP Request, etc. — not Webhook or Trigger) with test data shaped like the upstream output, then list_action_events on downstream actions to verify events flowed correctly
6. Document: Use create_note to add storyboard notes explaining the story's purpose, key design decisions, and what each branch handles. Position notes near the relevant actions on the diagram using position_x/position_y

Common Patterns

Webhook -> Transform -> HTTP Request

1. create_action(type="Agents::WebhookAgent", options_json="{\"path\":\"my-webhook\",\"verbs\":\"post\"}")
2. create_action(type="Agents::EventTransformationAgent", options_json="{\"mode\":\"message_only\",\"payload\":{...}}", source_ids_json="[\"<webhook_id>\"]")
3. create_action(type="Agents::HTTPRequestAgent", options_json="{\"url\":\"...\",\"method\":\"post\"}", source_ids_json="[\"<transform_id>\"]")

Webhook -> Condition -> Branch

1. create_action(type="Agents::TriggerAgent", name="Is Critical?",
                 options_json="{\"rules\":[{\"type\":\"field==value\",\"path\":\"...\",\"value\":\"critical\"}]}",
                 source_ids_json="[\"<webhook_id>\"]")
2. create_action(type="Agents::EventTransformationAgent", name="On Match", ...,
                 source_ids_json="[\"<trigger_id>\"]")
3. create_action(type="Agents::EventTransformationAgent", name="On No Match", ...,
                 links_to_sources_json="[{\"source_id\":\"<trigger_id>\",\"type\":\"NO_MATCH\"}]")

Error Handling

• Set emit_failure_event: "Always" on HTTP requests
• Use FAILURE link type to route errors to notification actions
• Use retry_on_status: ["429", "500-599"] for transient failures

Testing a Story

run_action works on executable action types (Event Transform, HTTP Request, Send to Story, etc.) but returns "Action type cannot be run" for Webhook and Trigger actions. To test a full pipeline, run the first executable action after the webhook with the expected upstream payload shape:

1. run_action(action_id="<first_executable_action_id>", payload_json="{\"body\":{\"test_key\":\"test_value\"}}")
   — use the first Event Transform, HTTP Request, or other executable action (not the Webhook/Trigger)
   — shape the payload to match what the upstream action would normally emit
2. list_action_events(action_id="<downstream_action_id>") to verify events flowed correctly
3. If an event failed, fix the action, then reemit_event(event_id="<failed_event_id>")

API Fallback

For operations not covered by the 31 MCP tools (e.g., managing story drafts, owners, recipients, cases, records, or admin operations), fall back to direct API calls. See api-reference.md for the full endpoint catalog with authentication and pagination details.