pipefy-introspection

Introspection & Raw GraphQL

Read only the reference for your active surface: MCP or CLI. The workflows below use shared operation names and arguments.

Schema discovery, organization info, and a fallback executor.

This is Tier 2 in the resolution strategy: when a dedicated operation fails or doesn't exist, use introspection to understand the API, then execute_graphql to run the operation directly.

If a dedicated write reports failure, re-read the target before trying a mutation through execute_graphql. Continue only when the read shows the write did not land. If the result is inconclusive, report it and stop; a second create can duplicate data.

Tier 1: dedicated operation exists — use it. Tier 2: use introspection + execute_graphql (this skill). Tier 3: direct curl/httpx fallback — see pipefy-api-fallback.

Tools

max_depth (introspect_type / query / mutation)

max_depth defaults to 1.

• `1` — type/field info only (no inlined sub-types).
• `2+` — resolves referenced input/output types inline (resolvedType), so one call can replace introspecting the mutation then each input type separately.

Example:

introspect_mutation mutation_name="createCard" max_depth=2

Scalars (ID, String, Int, …) are never expanded.

kind on search_schema

Optional filter: OBJECT, INPUT_OBJECT, ENUM, SCALAR, INTERFACE, UNION.

search_schema keyword="automation" kind="INPUT_OBJECT"

When to use introspection

• A dedicated tool returned an error and you need to understand why — introspect the input type to check argument names/types.
• Field types for create_phase_field: the tool's input schema lists them. When the schema is not visible, or to check for drift, introspect_type('FieldTypeId') lists them.
• Before using extra_input: introspect the corresponding input type to find optional keys.
• Unknown mutation signature: introspect_mutation('createSomething') before execute_graphql.
• Schema exploration: search_schema('automation') to find related types and inputs.

When to use execute_graphql

• No dedicated operation exists for the operation.
• A dedicated tool failed and you've used introspection to understand the correct payload.
• Ad-hoc queries like resolving an org UUID via pipe(id: $id) { organization { uuid } }.
• Complex nested queries that no single tool covers.

Always prefer dedicated operations. They validate inputs, handle pagination, and format errors consistently. execute_graphql is the fallback when dedicated tools can't solve the problem.

Steps — discover a mutation signature

1. Search for the mutation by keyword:

   search_schema keyword="label"

1. Get the full mutation signature:

   introspect_mutation mutation_name="createLabel"

1. Discover input type fields:

   introspect_type type_name="CreateLabelInput"

1. Execute execute_graphql with the discovered mutation and its variables.

Common fallback recipes

Ready-to-use patterns for situations where dedicated tools are insufficient.

Recipe 1 — Discover valid field types for create_phase_field

introspect_type('FieldTypeId')

CreatePhaseFieldInput.type is an ID scalar, but the API publishes its valid values as the FieldTypeId enum. The create_phase_field input schema lists the same values.

Recipe 2 — Find a card by title (not possible with find_cards)

find_cards only searches custom field values. To search by title:

execute_graphql query='query($pipeId: ID!, $first: Int) { cards(pipe_id: $pipeId, first: $first) { edges { node { id title current_phase { name } } } } }' variables='{"pipeId":"<pipe-id>","first":50}'

Filter by title client-side. For large pipes, paginate with after.

Recipe 3 — Discover what extra_input accepts for any mutation

When a tool accepts extra_input (e.g. create_automation, update_label), discover all optional keys:

introspect_mutation('createAutomation')     # find the input type name
introspect_type('CreateAutomationInput')    # see all inputFields

Compare with the tool's primary arguments to know which keys are additive via extra_input.

Recipe 4 — Discover organization IDs

To answer "which organizations do I have access to?" with nothing in hand, call list_organizations — it needs no id and returns each org's id, uuid, name, and your role. That is the entry point; reach for the GraphQL fallbacks below only when you already have a pipe.

When the user only has a pipe ID and needs its organization_id:

execute_graphql query='query($id: ID!) { pipe(id: $id) { organization { id uuid name } } }' variables='{"id":"<pipe-id>"}'

Recipe 5 — Update a select field's options after creation

To set options at creation, pass options to create_phase_field. Use this recipe to change them later.

execute_graphql query='mutation($id: ID!, $options: [String!]) { updatePhaseField(input: { id: $id, options: $options }) { phase_field { id label options } } }' variables='{"id":"<field-id>","options":["High","Medium","Low"]}'

Recipe 6 — Check phase transition rules

When move_card_to_phase fails with "not a valid target phase":

execute_graphql query='query($id: ID!) { phase(id: $id) { id name cards_can_be_moved_to_phases { id name } } }' variables='{"id":"<current-phase-id>"}'

Returns the valid destination phases from the current phase.

Success criteria

• introspect_type returns the complete field list for the input type.
• execute_graphql returns the expected data without errors.

Failure modes

• `introspect_type` returns `null`: type name is case-sensitive; try PascalCase (e.g., CreateLabelInput, not create_label_input).
• `search_schema` returns many hits: case-insensitive substring matching; broad keywords like "card" flood results. Prefer specific names like "AiAgent", "FieldCondition".
• `introspect_mutation` is expensive: fetches all root mutation fields and filters client-side (single large query). Prefer introspect_type on the specific input type when you already know the mutation name.
• `execute_graphql` returns GraphQL errors: check path and message.
• Raw HTTP endpoint: the toolkit sends public-schema introspection and operations to the same endpoint, <PIPEFY_BASE_URL>/graphql (default https://app.pipefy.com/graphql). For direct calls, see pipefy-api-fallback.

See also

• docs/mcp/tools/introspection.md — MCP parameters, query/mutation mismatch hints on execute_graphql.
• pipefy-api-fallback — Tier 3: direct HTTP fallback when the toolkit is unavailable.
• pipefy-pipes-and-cards — most common dedicated tools (prefer over execute_graphql).