Configure putior MCP Server

Set up the putior MCP server so AI assistants (Claude Code, Claude Desktop) can directly call workflow annotation and diagram generation tools.

When to Use

• Enabling AI assistants to interactively annotate and visualize workflows
• Setting up a new development environment with putior MCP integration
• After installing putior and wanting AI-assisted workflow documentation
• Configuring agent-to-agent communication via ACP for automated pipelines

Inputs

• Required: putior installed (see install-putior)
• Required: Target client: Claude Code, Claude Desktop, or both
• Optional: Whether to also configure ACP server (default: no)
• Optional: Custom host/port for ACP server (default: localhost:8080)

Procedure

Step 1: Install MCP Dependencies

Install the required packages for MCP server functionality.

# Required: MCP framework
remotes::install_github("posit-dev/mcptools")

# Required: Tool definition framework
install.packages("ellmer")

# Verify both load
library(mcptools)
library(ellmer)

Expected: Both packages install and load without errors.

On failure: mcptools requires remotes package. Install it first: install.packages("remotes"). If GitHub rate-limits, configure a GITHUB_PAT in ~/.Renviron (add the line GITHUB_PAT=your_token_here and restart R). Do not paste tokens into shell commands or commit them to version control.

Step 2: Configure Claude Code (WSL/Linux/macOS)

Add the putior MCP server to Claude Code's configuration.

# One-line setup
claude mcp add putior -- Rscript -e "putior::putior_mcp_server()"

For WSL with Windows R:

claude mcp add putior -- "/mnt/c/Program Files/R/R-4.5.2/bin/Rscript.exe" -e "putior::putior_mcp_server()"

Verify the configuration:

claude mcp list
claude mcp get putior

Expected: putior appears in the MCP server list with status "configured".

On failure: If Claude Code is not in PATH, add it: export PATH="$HOME/.claude/local/node_modules/.bin:$PATH". If the Rscript path is wrong, locate R with which Rscript or ls "/mnt/c/Program Files/R/".

Step 3: Configure Claude Desktop (Windows)

Add putior to Claude Desktop's MCP configuration file.

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "putior": {
      "command": "C:\\PROGRA~1\\R\\R-45~1.0\\bin\\x64\\Rscript.exe",
      "args": ["-e", "putior::putior_mcp_server()"]
    }
  }
}

Or with the full path:

{
  "mcpServers": {
    "putior": {
      "command": "C:\\Program Files\\R\\R-4.5.2\\bin\\x64\\Rscript.exe",
      "args": ["-e", "putior::putior_mcp_server()"]
    }
  }
}

Restart Claude Desktop after editing the configuration.

Expected: Claude Desktop shows putior in its MCP server list. Tools become available in conversation.

On failure: Validate JSON syntax with a JSON linter. Check that the R path exists. Use 8.3 short names (PROGRA~1, R-45~1.0) if spaces in paths cause issues.

Step 4: Verify All 16 Tools

Test that all MCP tools are accessible and functional.

# Get tool definitions
tools <- putior::putior_mcp_tools()
cat(sprintf("Total tools: %d\n", length(tools)))

# List tool names
vapply(tools, function(t) t$name, character(1))

The 16 tools organized by category:

Core Workflow (5):

• put — Scan files for PUT annotations (supports exclude parameter for regex-based file filtering)
• put_diagram — Generate Mermaid diagrams
• put_auto — Auto-detect workflow from code (supports exclude parameter)
• put_generate — Generate annotation suggestions (supports exclude parameter)
• put_merge — Merge manual + auto annotations (supports exclude parameter)

Reference/Discovery (7):

• get_comment_prefix — Get comment prefix for extension
• get_supported_extensions — List supported extensions
• list_supported_languages — List supported languages
• get_detection_patterns — Get auto-detection patterns
• get_diagram_themes — List available themes
• putior_skills — AI assistant documentation
• putior_help — Quick reference help

Utilities (3):

• is_valid_put_annotation — Validate annotation syntax
• split_file_list — Parse file lists
• ext_to_language — Extension to language name

Configuration (1):

• set_putior_log_level — Configure logging verbosity

Test core tools from Claude Code:

Use the putior_help tool to see available commands
Use the put tool to scan ./R/ for annotations
Use the put_diagram tool to generate a diagram

Expected: All 16 tools listed. Core tools return expected results when called with valid inputs.

On failure: If tools are missing, check that putior version is current: packageVersion("putior"). Older versions may have fewer tools. Update with remotes::install_github("pjt222/putior").

Step 5: Configure ACP Server (Optional)

Set up the ACP (Agent Communication Protocol) server for agent-to-agent communication.

# Install ACP dependency
install.packages("plumber2")

# Start ACP server (blocks — run in a separate R session or background)
putior::putior_acp_server()

# Custom host/port
putior::putior_acp_server(host = "0.0.0.0", port = 9000)

Test ACP endpoints:

# Discover agent
curl http://localhost:8080/agents

# Execute a scan
curl -X POST http://localhost:8080/runs \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "parts": [{"content": "scan ./R/"}]}]}'

# Generate diagram
curl -X POST http://localhost:8080/runs \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "parts": [{"content": "generate diagram for ./R/"}]}]}'

Expected: ACP server starts on the configured port. /agents returns the putior agent manifest. /runs accepts natural language requests and returns workflow results.

On failure: If port 8080 is in use, specify a different port. If plumber2 is not installed, the server function will print a helpful error message suggesting installation.

Validation

• putior::putior_mcp_tools() exposes the core tools (put, put_diagram, put_auto, put_generate, put_merge) and returns ~16 tools for the current version
• Claude Code: claude mcp list shows putior configured
• Claude Code: putior_help tool returns help text when invoked
• Claude Desktop: putior appears in the MCP server list after restart
• Core tools (put, put_diagram, put_auto) execute without errors
• (Optional) ACP server responds to curl http://localhost:8080/agents

Common Pitfalls

• mcptools not installed: The MCP server requires mcptools (from GitHub) and ellmer (from CRAN). Both must be installed. putior checks and provides helpful messages if they're missing.
• Wrong R path in Claude Desktop: Windows paths need escaping in JSON (\\). Use 8.3 short names to avoid spaces: C:\\PROGRA~1\\R\\R-45~1.0\\bin\\x64\\Rscript.exe.
• Forgetting to restart: Claude Desktop must be restarted after editing the config file. Claude Code picks up changes on next session start.
• renv isolation: If putior is installed in an renv library but Claude Code/Desktop launches R without renv, the packages won't be found. Ensure mcptools and ellmer are installed in the global library or configure renv activation in the MCP server command.
• Port conflicts for ACP: The default ACP port (8080) is commonly used. Check with lsof -i :8080 or netstat -tlnp | grep 8080 before starting.
• Including only specific tools: To expose a subset of tools, use putior_mcp_tools(include = c("put", "put_diagram")) when building custom MCP server wrappers.
• Custom palettes via MCP: The palette parameter on put_diagram requires a putior_theme R object (created by put_theme()), which cannot be serialized through MCP's JSON interface. Use the built-in theme parameter string for MCP calls. For custom palettes, use R directly.

Related Skills

• install-putior — prerequisite: putior and optional deps must be installed
• configure-mcp-server — general MCP server configuration for Claude Code/Desktop
• troubleshoot-mcp-connection — diagnose connection issues if tools don't appear
• build-custom-mcp-server — build custom MCP servers that wrap putior tools
• analyze-codebase-workflow — use MCP tools interactively for codebase analysis