forge
Clone forge-starter
plane: controlresource: EvalRunC30rendered from forge v1.56.0 platform-model.json

eval

Drive real models as MCP clients against a suite; grade with asserts and a judge; costs real money.

EvalPOST /capabilities/eval
EvalRunthe Resource it produces
1 eventEvalRunCompleted

Ask your agent forge → eval

you
Run the core eval suite against hello's MCP server.
claude
forge → eval Done — EvalRun created; I recorded EvalRunCompleted.

From the terminal

$ forge eval suites/core.json --app hello --mcp-url https://hello.example.com/mcp

HTTP :3717

POST /capabilities/eval
Content-Type: application/json
x-forge-actor-type: builder
x-forge-actor-id: cli

{
  "app": "hello",
  "suite": "suites/core.json",
  "mcp_url": "https://hello.example.com/mcp"
}

Input fields 7

app
suite
mcp_url
models
judge_model
run_name
langfuse

Events ForgeEvent

EvalRunCompleted   resource_type: EvalRun

Resource GET /resources/EvalRun/:id

{
  "type": "EvalRun",
  "id": "res_01J…",
  "app_id": "app_hello",
  "status": "succeeded",
  "created_at": "2026-09-03T14:00:00Z"
}

Errors: see Reference → Errors. Find it in the playground