Utiliser mcpc pour interagir avec les serveurs MCP

Apprenez à utiliser mcpc pour appeler des outils, lire des ressources et accéder aux serveurs MCP depuis la ligne de commande. Plus efficace que les appels de fonction directe.

Spar Skills Guide Bot
DeveloppementIntermédiaire
15011/03/2026
Claude CodeCursorCopilot
#mcp-protocol#cli-tools#bash-scripting#model-context-protocol#command-line

name: using-mcp-tools-with-mcpc description: Use mcpc CLI to interact with MCP servers - call tools, read resources, get prompts. Use when working with Model Context Protocol servers, calling MCP tools, or accessing MCP resources programmatically; prefer key:=value bindings over raw JSON bodies. allowed-tools: Bash(mcpc:) Bash(node dist/cli/index.js:) Read Grep

mcpc: MCP command-line client

Use mcpc to interact with MCP (Model Context Protocol) servers from the command line. This is more efficient than function calling - generate shell commands instead.

Trust pattern

  • Always: Read-only mcpc commands in the sandbox (e.g., tools-list, tools-get, resources-list/read, prompts-list/get, tools-call for read/search-only endpoints), session status checks, and commands that reuse already-created auth profiles.
  • Ask: Anything that writes or needs network/OAuth (login/logout must be human-initiated in the foreground), connecting to new servers, commands that create/update/delete data, helper scripts that write files, or when sandbox blocks the command.
  • Never: Destructive workspace actions (moves/deletes) without explicit user request; connecting to unknown MCP servers without instruction; backgrounding mcpc <server> login or trying to auto-open a browser.

Quick reference

# List sessions and auth profiles
mcpc

# Show server info
mcpc <server>
mcpc @<session>

# Tools
mcpc <target> tools-list
mcpc <target> tools-get <tool-name>
mcpc <target> tools-call <tool-name> key:=value key2:="string value"

# Resources
mcpc <target> resources-list
mcpc <target> resources-read <uri>

# Prompts
mcpc <target> prompts-list
mcpc <target> prompts-get <prompt-name> arg1:=value1

# Sessions (persistent connections)
mcpc <server> connect @<name>
mcpc @<name> <command>
mcpc @<name> close

# Authentication
mcpc <server> login
mcpc <server> logout

Target types

  • mcp.example.com - Direct HTTPS connection to remote server
  • localhost:8080 or 127.0.0.1:8080 - Local HTTP server (http:// is default for localhost)
  • @session-name - Named persistent session (faster, maintains state)
  • config-entry - Entry from config file (with --config)

Passing arguments

Prefer key:=value bindings. Use inline JSON only when needed (e.g., first-arg object or complex arrays):

# String values
mcpc @s tools-call search query:="hello world"

# Numbers, booleans, null (auto-parsed as JSON)
mcpc @s tools-call search query:="hello" limit:=10 enabled:=true

# Complex JSON values
mcpc @s tools-call search config:='{"nested":"value"}' items:='[1,2,3]'

# Force string type with JSON quotes
mcpc @s tools-call search id:='"123"'

# Inline JSON object (if first arg starts with { or [)
mcpc @s tools-call search '{"query":"hello","limit":10}'

# From stdin (auto-detected when piped)
echo '{"query":"hello"}' | mcpc @s tools-call search

JSON output for scripting

Always use --json flag for machine-readable output:

# Get tools as JSON
mcpc --json @apify tools-list

# Call tool and parse result with jq
mcpc --json @apify tools-call search query:="test" | jq '.content[0].text'

# Chain commands
mcpc --json @server1 tools-call get-data | mcpc @server2 tools-call process

Sessions for efficiency

Create sessions for repeated interactions:

# Create session (or reconnect if exists)
mcpc mcp.apify.com connect @apify

# Use session (faster - no reconnection overhead)
mcpc @apify tools-list
mcpc @apify tools-call search query:="test"

# Restart session (useful after server updates)
mcpc @apify restart

# Close when done
mcpc @apify close

Session states:

  • 🟢 live - Bridge running, server might or might not be responding
  • 🟡 crashed - Bridge crashed; auto-restarts on next use
  • 🔴 expired - Server rejected session; needs close and reconnect

Authentication

OAuth (interactive login – human-only, foreground):

  • mcpc <server> login opens the browser; mcpc never opens it itself. Do not background this command or it will miss the localhost callback.
  • Run login once per profile, then reuse the saved credentials in scripts.
  • Re-run login to refresh/change scopes.

Python preflight to enforce “login first” in scripts (no automatic browser launches):

import json, os, sys

server = os.environ.get("MCP_SERVER", "mcp.apify.com")
profile = os.environ.get("MCP_PROFILE", "default")
profiles_path = os.path.join(os.path.expanduser("~"), ".mcpc", "profiles.json")

try:
    data = json.load(open(profiles_path, "r", encoding="utf-8"))
    profiles = data.get("profiles", [])
except FileNotFoundError:
    profiles = []

has_profile = any(p.get("server") == server and p.get("name") == profile for p in profiles)
if not has_profile:
    print(f"No mcpc auth profile '{profile}' for {server}.")
    print(f"Run this yourself (foreground): mcpc {server} login --profile {profile}")
    sys.exit(1)

After the preflight succeeds, scripts may call mcpc --profile <name> ... or rely on the default profile.

Bearer token:

mcpc -H "Authorization: Bearer $TOKEN" mcp.apify.com tools-list
mcpc -H "Authorization: Bearer $TOKEN" mcp.apify.com connect @myserver

Proxy server for AI isolation

Create a proxy MCP server that hides authentication tokens:

# Human creates authenticated session with proxy
mcpc mcp.apify.com connect @ai-proxy --proxy 8080

# AI agent connects to proxy (no access to original tokens)
# Note: localhost defaults to http://
mcpc localhost:8080 tools-list
mcpc 127.0.0.1:8080 connect @sandboxed

Common patterns

List and inspect tools:

mcpc @s tools-list
mcpc @s tools-get tool-name

Call tool and extract text result:

mcpc --json @s tools-call my-tool | jq -r '.content[0].text'

Read resource content:

mcpc @s resources-read "file:///path/to/file"

Use config file for local servers:

mcpc --config .vscode/mcp.json filesystem resources-list

Exit codes

  • 0 - Success
  • 1 - Client error (invalid arguments)
  • 2 - Server error (tool failed)
  • 3 - Network error
  • 4 - Authentication error

Debugging

# Verbose output shows protocol details
mcpc --verbose @s tools-call my-tool

Example script

See docs/examples/company-lookup.sh for a complete example of an AI-generated script that validates prerequisites and calls MCP tools.

Skills similaires