APICP · OpenAPI to MCP

Turn your API into MCP tools.

APICP turns any OpenAPI spec into a focused MCP server — the right operations exposed, auth handled, responses cleaned. Your Frags Blueprints call it like any other tool. No bespoke MCP server to build or maintain.

openapi.yaml → apicp.yaml → MCP server

Your OpenAPI spec

  • GET /plans
  • POST /plans
  • GET /plans/{id}
  • PUT /plans/{id}
  • DELETE /plans/{id}
  • GET /users
  • POST /webhooks
  • GET /audit-logs

One MCP tool

searchPlans

Search for blueprints by partial name match.

Model sets
search
Config
Authorization header
Returns
[*].{id, name}
apicp.yaml
Copy
context: {} server:  url: https://api.example.com/v2 mappers:  - name: searchPlans    description: Search for blueprints by partial name match.    values:      - path: header.Authorization        value: "Bearer {{ .token }}"    inputSchema:      search:        path: query.search        description: Partial Blueprint name to search for.    outputTransformer: "[*].{id: id, name: name}"

Why it exists

API access that matches how agent workflows actually run.

Most OpenAPI specs are written for client libraries and human developers. APICP makes the same APIs legible to models by narrowing scope, stabilizing inputs, and cleaning responses before they enter context.

Select the operations worth exposing

Define one mapper per MCP tool so the model sees a small, intentional API surface instead of an entire OpenAPI catalog.

Shape inputs for the model

Move auth, tenancy, defaults, and server overrides into config while keeping only the right fields in the LLM-facing schema.

Return clean context

Use JMESPath transformers to trim noisy API payloads into the exact fields a workflow needs for reliable downstream reasoning.

How it works

From spec to MCP server in minutes.

APICP is the on-ramp between the APIs you already have and the deterministic Blueprints you run on Frags. Bring a spec; leave with a tool your workflows can call.

  1. Point it at your OpenAPI spec

    Bring any REST API's spec. No SDK to wrap, no bespoke MCP server to write and keep in sync.

  2. Map what matters, in YAML

    Pick the operations worth exposing, move auth and tenancy into config, and trim noisy responses with JMESPath — all declarative.

  3. Frags Blueprints call it as a tool

    The generated MCP drops straight into your Blueprints. Frags calls it like any other tool — deterministic and schema-validated, on your infra or Barndoor Automations cloud.

Config surface

Small primitives, explicit control.

APICP keeps API mapping declarative. You decide which request fields are fixed, which ones the model supplies, and how response data is shaped before it's handed back to the workflow.

  • contextInbound headers and shared variables
  • valuesHardcoded or templated request fields
  • server.urlA clean base URL when the spec is wrong
  • mappersThe MCP tools APICP exposes
  • inputSchemaLLM-facing inputs mapped to request paths
  • outputTransformerJMESPath cleanup before the model sees data

APICP and Frags work together.

APICP narrows the API into clean tools. Frags runs the Blueprint that decides when they're called and what happens to the result.