Plumber API MCP支持工具

# Install from GitHub
remotes::install_github("armish/plumber2mcp")

library(plumber)
library(plumber2mcp)

# Create and run a Plumber API with MCP support via HTTP
pr("plumber.R") %>%
  pr_mcp(transport = "http") %>%
  pr_run(port = 8000)

library(plumber)
library(plumber2mcp)

# Create and run a Plumber API with native stdio transport
pr("plumber.R") %>%
  pr_mcp(transport = "stdio")

{
  "mcpServers": {
    "plumber2mcp": {
      "command": "Rscript",
      "args": ["-e", "plumber::pr('api.R') %>% plumber2mcp::pr_mcp(transport='stdio')"],
      "cwd": "."
    }
  }
}

mcp-cli servers  # Should show your server as "Ready"
mcp-cli tools    # List available tools
mcp-cli cmd --tool GET__echo --tool-args '{"msg": "Hello!"}'  # Call a tool

#* Calculate statistical operations on numeric data
#* 
#* This endpoint performs various statistical calculations on a vector of numbers.
#* It supports multiple operations and handles missing values.
#* 
#* @param numbers Numeric vector of values to calculate statistics for
#* @param operation Statistical operation to perform: "mean", "median", "sum", "sd" (default: "mean")
#* @param na_rm:bool Logical value indicating whether to remove NA values (default: TRUE)
#* @param digits:int Number of decimal places to round the result (default: 2)
#* @return List containing the calculated result and metadata
#* @post /calculate
function(numbers, operation = "mean", na_rm = TRUE, digits = 2) {
  # Convert input to numeric
  if (is.character(numbers)) {
    numbers <- as.numeric(strsplit(numbers, ",")[[1]])
  } else {
    numbers <- as.numeric(numbers)
  }
  
  result <- switch(operation,
    "mean" = mean(numbers, na.rm = na_rm),
    "median" = median(numbers, na.rm = na_rm),
    "sum" = sum(numbers, na.rm = na_rm),
    "sd" = sd(numbers, na.rm = na_rm),
    stop("Unknown operation: ", operation)
  )
  
  list(
    result = round(result, digits),
    operation = operation,
    count = length(numbers[!is.na(numbers)]),
    input_length = length(numbers)
  )
}

Calculate statistical operations on numeric data

This endpoint performs various statistical calculations on a vector of numbers.
It supports multiple operations and handles missing values.

Parameters:
- numbers (string): Numeric vector of values to calculate statistics for
- operation (string) [default: mean]: Statistical operation to perform: "mean", "median", "sum", "sd"
- na_rm (boolean) [default: TRUE]: Logical value indicating whether to remove NA values
- digits (integer) [default: 2]: Number of decimal places to round the result

HTTP Method: POST
Path: /calculate

{
  "type": "object",
  "properties": {
    "numbers": {
      "type": "string",
      "description": "Numeric vector of values to calculate statistics for"
    },
    "operation": {
      "type": "string",
      "description": "Statistical operation to perform: \"mean\", \"median\", \"sum\", \"sd\"",
      "default": "mean"
    },
    "na_rm": {
      "type": "boolean",
      "description": "Logical value indicating whether to remove NA values",
      "default": true
    },
    "digits": {
      "type": "integer",
      "description": "Number of decimal places to round the result",
      "default": 2
    }
  },
  "required": ["numbers"]
}

{
  "type": "object",
  "description": "Structured response object",
  "properties": {
    "result": {
      "type": "number",
      "description": "Response field: result"
    },
    "operation": {
      "type": "string",
      "description": "Response field: operation"
    },
    "count": {
      "type": "number",
      "description": "Response field: count"
    },
    "input_length": {
      "type": "number",
      "description": "Response field: input_length"
    }
  }
}

User: "Can you help me calculate the mean of some numbers using this API?"
AI: "I can see there's a calculate endpoint, but I'm not sure what parameters it needs..."

User: "Can you help me calculate the mean of some numbers using this API?"
AI: "I can see you have a calculate endpoint that performs statistical operations! 
     It needs a 'numbers' parameter (required) and optionally 'operation' (defaults to 'mean'), 
     'na_rm' (defaults to TRUE), and 'digits' (defaults to 2). 
     
     Let me call it for you:
     POST /calculate with {"numbers": "1,2,3,4,5"}
     
     This will return an object with the result, operation used, count of valid numbers, 
     and input length."

#* Echo back the input
#* 
#* Simple endpoint that echoes back whatever message you send it
#* 
#* @param msg:string The message to echo back (default: "Hello World")
#* @get /echo
function(msg = "Hello World") {
  list(message = paste("Echo:", msg))
}

#* Add two numbers together
#* 
#* Performs arithmetic addition on two numeric values
#* 
#* @param a:number First number to add
#* @param b:number Second number to add  
#* @param precision:int Number of decimal places to round result (default: 2)
#* @post /add
function(a, b, precision = 2) {
  result <- as.numeric(a) + as.numeric(b)
  list(
    result = round(result, precision),
    operation = "addition",
    inputs = c(a, b)
  )
}

# Simple prompt with no arguments
pr %>%
  pr_mcp(transport = "stdio") %>%
  pr_mcp_prompt(
    name = "r-help",
    description = "Get help with R programming",
    func = function() {
      paste(
        "I need help with R programming.",
        "Please provide guidance on best practices and common patterns.",
        sep = "\n"
      )
    }
  )

# Prompt with arguments
pr %>%
  pr_mcp(transport = "stdio") %>%
  pr_mcp_prompt(
    name = "analyze-dataset",
    description = "Generate a comprehensive analysis plan for an R dataset",
    arguments = list(
      list(
        name = "dataset",
        description = "Name of the R dataset to analyze",
        required = TRUE
      ),
      list(
        name = "focus",
        description = "Specific aspect to focus on",
        required = FALSE
      )
    ),
    func = function(dataset, focus = "general") {
      sprintf(
        paste(
          "Please analyze the %s dataset in R.",
          "Focus: %s",
          "",
          "Provide:",
          "1. Summary statistics",
          "2. Data quality assessment",
          "3. Key insights",
          sep = "\n"
        ),
        dataset, focus
      )
    }
  )

# Multi-turn conversation prompt
pr %>%
  pr_mcp(transport = "stdio") %>%
  pr_mcp_prompt(
    name = "code-review",
    description = "Review R code for quality and best practices",
    arguments = list(
      list(name = "code", description = "The R code to review", required = TRUE)
    ),
    func = function(code) {
      list(
        list(
          role = "user",
          content = list(
            type = "text",
            text = paste("Please review this R code:", code, sep = "\n\n")
          )
        ),
        list(
          role = "assistant",
          content = list(
            type = "text",
            text = "I'll review your code for correctness, performance, and style."
          )
        ),
        list(
          role = "user",
          content = list(
            type = "text",
            text = "Please provide specific suggestions for improvement."
          )
        )
      )
    }
  )

func = function() "Hello World"

func = function() {
  list(
    role = "user",
    content = list(type = "text", text = "Your message")
  )
}

func = function() {
  list(
    list(role = "user", content = list(type = "text", text = "First message")),
    list(role = "assistant", content = list(type = "text", text = "Second message"))
  )
}

pr_mcp_prompt(
  pr,
  name = "data-pipeline",
  description = "Guide for building data processing pipelines",
  arguments = list(
    list(name = "data_type", description = "Type of data to process", required = TRUE)
  ),
  func = function(data_type) {
    sprintf("Create a data processing pipeline for %s data...", data_type)
  }
)

pr_mcp_prompt(
  pr,
  name = "create-endpoint",
  description = "Template for creating new Plumber endpoints",
  func = function() {
    paste(
      "Generate a Plumber endpoint with:",
      "1. Proper roxygen documentation",
      "2. Input validation",
      "3. Error handling",
      "4. Example usage",
      sep = "\n"
    )
  }
)

pr_mcp_prompt(
  pr,
  name = "statistical-test",
  description = "Choose and implement appropriate statistical tests",
  arguments = list(
    list(name = "research_question", description = "Research question", required = TRUE)
  ),
  func = function(research_question) {
    sprintf(
      paste(
        "Research Question: %s",
        "",
        "Help me:",
        "1. Choose the appropriate statistical test",
        "2. Check assumptions",
        "3. Implement in R",
        "4. Interpret results",
        sep = "\n"
      ),
      research_question
    )
  }
)

library(plumber)
library(plumber2mcp)

pr("api.R") %>%
  pr_mcp(transport = "stdio") %>%

  # Add analysis prompt
  pr_mcp_prompt(
    name = "analyze",
    description = "Analyze data from the API",
    func = function() {
      "Guide me through analyzing the data available from this API."
    }
  ) %>%

  # Add troubleshooting prompt
  pr_mcp_prompt(
    name = "troubleshoot",
    description = "Help troubleshoot API issues",
    arguments = list(
      list(name = "issue", description = "Description of the issue", required = TRUE)
    ),
    func = function(issue) {
      sprintf("I'm experiencing this issue with the API: %s\n\nHow can I resolve it?", issue)
    }
  )

# Create a Plumber API with resources
pr(...) %>%
  pr_mcp(transport = "stdio") %>%
  
  # Add a resource that provides dataset information
  pr_mcp_resource(
    uri = "/data/iris-summary",
    func = function() {
      paste(
        "Dataset: iris",
        paste("Dimensions:", paste(dim(iris), collapse = " x ")),
        "",
        capture.output(summary(iris)),
        sep = "\n"
      )
    },
    name = "Iris Dataset Summary",
    description = "Statistical summary and structure of the iris dataset"
  ) %>%
  
  # Add a resource that shows current memory usage
  pr_mcp_resource(
    uri = "/system/memory",
    func = function() {
      mem <- gc()
      paste(
        "R Memory Usage:",
        paste("Used (Mb):", round(sum(mem[,2]), 2)),
        paste("Gc Trigger (Mb):", round(sum(mem[,6]), 2)),
        "",
        "Object sizes in workspace:",
        capture.output(print(object.size(ls.str()), units = "Mb")),
        sep = "\n"
      )
    },
    name = "R Memory Usage",
    description = "Current R session memory statistics"
  ) %>%
  
  # Add a resource with model diagnostics
  pr_mcp_resource(
    uri = "/models/latest-lm",
    func = function() {
      # Example: fit a model and return diagnostics
      model <- lm(mpg ~ wt + hp, data = mtcars)
      paste(
        "Linear Model: mpg ~ wt + hp",
        "",
        capture.output(summary(model)),
        "",
        "Diagnostic plots saved to: /tmp/lm_diagnostics.png",
        sep = "\n"
      )
    },
    name = "Latest Linear Model",
    description = "Diagnostics for the most recent linear model",
    mimeType = "text/plain"
  )

pr(...) %>% 
  pr_mcp(transport = "stdio") %>%
  pr_mcp_help_resources()  # Adds help for common R functions

# Create resources based on available data files
data_files <- list.files("data/", pattern = "\\.csv$")

my_pr <- pr(...)
for (file in data_files) {
  my_pr <- my_pr %>%
    pr_mcp_resource(
      uri = paste0("/data/", tools::file_path_sans_ext(file)),
      func = local({
        current_file <- file  # Capture the file name
        function() {
          data <- read.csv(file.path("data", current_file))
          paste(
            paste("File:", current_file),
            paste("Rows:", nrow(data)),
            paste("Columns:", paste(names(data), collapse = ", ")),
            "",
            "First 5 rows:",
            capture.output(print(head(data, 5))),
            sep = "\n"
          )
        }
      }),
      name = paste("Dataset:", tools::file_path_sans_ext(file)),
      description = paste("Contents of", file)
    )
}

You: "What datasets are available in this R session?"
AI: *Lists resources and reads relevant ones*
    "I can see you have the iris dataset available. Based on the 
    /data/iris-summary resource, it has 150 observations of 5 variables..."

You: "Can you help me analyze the relationship between Sepal.Length and Petal.Length?"
AI: *Reads the iris summary resource for context*
    "Based on the iris dataset summary, I can see that Sepal.Length ranges 
    from 4.3 to 7.9 and Petal.Length from 1.0 to 6.9. Let me create a 
    linear model to analyze their relationship..."

my_pr %>% pr_mcp(transport = "http", path = "/my-mcp-server")

# Include only specific endpoints
my_pr %>% pr_mcp(transport = "http", include_endpoints = c("GET__echo", "POST__add"))

# Exclude specific endpoints
my_pr %>% pr_mcp(transport = "stdio", exclude_endpoints = c("POST__internal"))

my_pr %>% pr_mcp(
  transport = "http",
  server_name = "my-api-mcp",
  server_version = "1.0.0"
)

#* @apiTitle My MCP-Enabled API
#* @apiDescription API with MCP support for AI assistants

#* Get current time
#* @get /time
function() {
  list(time = Sys.time())
}

#* Calculate factorial
#* @param n Integer to calculate factorial
#* @post /factorial
function(n) {
  n <- as.integer(n)
  if (n < 0) stop("n must be non-negative")
  list(result = factorial(n))
}

library(plumber)
library(plumber2mcp)

pr("my_api.R") %>%
  pr_mcp(transport = "http") %>%
  pr_run(port = 8000)

source(system.file("examples/run_mcp_server.R", package = "plumber2mcp"))

source(system.file("examples/test_mcp_client.R", package = "plumber2mcp"))

{
  "mcpServers": {
    "my-r-api": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

library(plumber)
library(plumber2mcp)

pr("api.R") %>%
  pr_mcp(transport = "http") %>%
  pr_run(port = 8000)

cd /path/to/plumber2mcp/inst/examples
mcp-inspector --config http_wrapper_config.json --server plumber2mcp

cd /path/to/plumber2mcp/inst/examples
mcp-inspector --config stdio_config.json --server plumber2mcp

# For stdio transport
my_pr %>% pr_mcp(transport = "stdio", debug = TRUE)

# For HTTP transport (debug not available)
my_pr %>% pr_mcp(transport = "http") %>% pr_run(port = 8000)