Tools Reference

Complete reference for all 7 RelayPlane MCP Server tools.

Tool	Purpose	Cost
relay_run	Single AI model call	Provider cost
relay_workflow_run	Multi-step workflows	Provider cost
relay_workflow_validate	DAG validation	Free
relay_models_list	List available models	Free
relay_skills_list	Discover skills	Free
relay_runs_list	Recent runs	Free
relay_run_get	Run details + trace	Free

relay_run

Execute a single AI model call. Useful for testing prompts before building full workflows.

Input Schema

1{
2  model: string,        // "provider:model" format (e.g., "openai:gpt-4o")
3  prompt: string,       // The user prompt to send
4  systemPrompt?: string, // Optional system prompt
5  schema?: object       // Optional JSON schema for structured output
6}

Response

1{
2  success: boolean,
3  output: string | object,
4  model: string,
5  usage: {
6    promptTokens: number,
7    completionTokens: number,
8    totalTokens: number,
9    estimatedProviderCostUsd: number
10  },
11  durationMs: number,
12  runId: string,
13  traceUrl: string,
14  error?: { code: string, message: string }
15}

Example

1relay_run({
2  model: "openai:gpt-5.2",
3  prompt: "Extract the company name from: john@acme.com",
4  schema: {
5    type: "object",
6    properties: { company: { type: "string" } },
7    required: ["company"]
8  }
9})

relay_workflow_run

Execute a multi-step AI workflow. Intermediate results stay in the workflow engine (not your context), providing 90%+ context reduction on complex pipelines.

Input Schema

1{
2  name: string,         // Workflow name for tracing
3  steps: Array<{
4    name: string,       // Step identifier
5    model?: string,     // "provider:model" format
6    prompt?: string,    // Prompt template (supports {{interpolation}})
7    systemPrompt?: string,
8    depends?: string[], // Dependencies on other steps
9    mcp?: string,       // MCP tool ("server:tool" format)
10    params?: object,    // MCP tool parameters
11    schema?: object     // JSON schema for structured output
12  }>,
13  input: object         // Input data (accessible via {{input.field}})
14}

Response

1{
2  success: boolean,
3  steps: Record
4    success: boolean,
5    output: any,
6    durationMs: number,
7    usage?: {
8      promptTokens: number,
9      completionTokens: number,
10      estimatedProviderCostUsd: number
11    },
12    error?: { code: string, message: string }
13  }>,
14  finalOutput: any,
15  totalUsage: {
16    totalTokens: number,
17    estimatedProviderCostUsd: number
18  },
19  totalDurationMs: number,
20  runId: string,
21  traceUrl: string,
22  contextReduction: string  // e.g., "94% (saved ~45k tokens)"
23}

Example

1relay_workflow_run({
2  name: "invoice-processor",
3  steps: [
4    {
5      name: "extract",
6      model: "openai:gpt-5.2",
7      prompt: "Extract invoice data from: {{input.fileContent}}"
8    },
9    {
10      name: "validate",
11      model: "anthropic:claude-sonnet-4.5",
12      depends: ["extract"],
13      prompt: "Verify totals match in: {{steps.extract.output}}"
14    },
15    {
16      name: "summarize",
17      model: "openai:gpt-5-nano",
18      depends: ["validate"],
19      prompt: "Create 2-sentence summary: {{steps.validate.output}}"
20    }
21  ],
22  input: { fileContent: "..." }
23})

relay_workflow_validate

Validate workflow structure without making any LLM calls. Free to use. Checks DAG structure (no cycles), dependency references, and model ID format.

Use this tool before relay_workflow_run to catch structural errors without spending on provider costs.

Input Schema

1{
2  steps: Array<{
3    name: string,
4    model?: string,
5    prompt?: string,
6    depends?: string[],
7    mcp?: string,
8    params?: object
9  }>
10}

Response

1{
2  valid: boolean,
3  errors: Array<{
4    step: string,
5    field: string,
6    message: string
7  }>,
8  warnings: Array<{
9    step: string,
10    message: string
11  }>,
12  structure: {
13    totalSteps: number,
14    executionOrder: string[],
15    parallelGroups: string[][]
16  }
17}

Validates

✓DAG structure (no cycles)
✓Dependency references exist
✓Model IDs are valid format
✓Required fields present

Does NOT Validate

✗Schema compatibility between steps
✗Prompt effectiveness

relay_models_list

List available AI models with capabilities and pricing. Use to check valid model IDs before testing.

Input Schema

1{
2  provider?: "openai" | "anthropic" | "google" | "xai"  // Optional filter
3}

Response

1{
2  models: Array<{
3    id: string,             // e.g., "openai:gpt-4o"
4    provider: string,
5    name: string,
6    capabilities: string[], // e.g., ["text", "vision", "function_calling"]
7    contextWindow: number,
8    inputCostPer1kTokens: number,
9    outputCostPer1kTokens: number
10  }>
11}

relay_skills_list

List available pre-built workflow skills. Skills are reusable patterns for common tasks with documented context reduction metrics.

Input Schema

1{
2  category?: "extraction" | "content" | "integration" | "all"
3}

Response

1{
2  skills: Array<{
3    name: string,
4    category: string,
5    description: string,
6    models: string[],
7    contextReduction: string,  // e.g., "97%"
8    usage: string              // Example usage
9  }>
10}

relay_runs_list

List recent workflow runs for debugging and reference.

Input Schema

1{
2  limit?: number  // Default: 10, max: 50
3}

Response

1{
2  runs: Array<{
3    runId: string,
4    name: string,
5    status: "success" | "error",
6    createdAt: string,
7    durationMs: number,
8    totalCost: number
9  }>
10}

relay_run_get

Get full details of a specific run including all step outputs and trace URL.

Input Schema

1{
2  runId: string  // The run ID to retrieve
3}

Response

1{
2  runId: string,
3  name: string,
4  status: "success" | "error",
5  steps: Record
6    output: any,
7    durationMs: number,
8    usage?: object
9  }>,
10  finalOutput: any,
11  traceUrl: string,
12  createdAt: string,
13  totalDurationMs: number
14}

Next Steps

Skills — Pre-built workflow patterns with context reduction metrics
Budget & Limits — Configure safety limits for provider costs