Use MCP servers on-demand via the mcp CLI tool - discover tools, resources, and prompts without polluting context with pre-loaded MCP integrations
This skill inherits all available tools. When active, it can use any tool Claude has access to.
Use the mcp CLI tool to dynamically discover and invoke MCP server capabilities without pre-configuring them as permanent integrations.
Use this skill when you need to:
The mcp CLI must be installed at ~/.local/bin/mcp. If not present:
# Clone and build
cd /tmp && git clone --depth 1 https://github.com/f/mcptools.git
cd mcptools && CGO_ENABLED=0 go build -o ~/.local/bin/mcp ./cmd/mcptools
Always ensure PATH includes the binary:
export PATH="$HOME/.local/bin:$PATH"
mcp tools <server-command>
Examples:
# Filesystem server
mcp tools npx -y @modelcontextprotocol/server-filesystem /path/to/allow
# Memory/knowledge graph server
mcp tools npx -y @modelcontextprotocol/server-memory
# GitHub server (requires token)
mcp tools docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server
# HTTP-based server
mcp tools https://example.com/mcp
mcp resources <server-command>
Resources are data sources the server exposes (files, database entries, etc.).
mcp prompts <server-command>
Prompts are pre-defined prompt templates the server provides.
# For full schema details including parameter types
mcp tools --format json <server-command>
mcp tools --format pretty <server-command>
mcp call <tool_name> --params '<json>' <server-command>
Read a file:
mcp call read_file --params '{"path": "/tmp/example.txt"}' \
npx -y @modelcontextprotocol/server-filesystem /tmp
Write a file:
mcp call write_file --params '{"path": "/tmp/test.txt", "content": "Hello world"}' \
npx -y @modelcontextprotocol/server-filesystem /tmp
List directory:
mcp call list_directory --params '{"path": "/tmp"}' \
npx -y @modelcontextprotocol/server-filesystem /tmp
Create entities (memory server):
mcp call create_entities --params '{"entities": [{"name": "Project", "entityType": "Software", "observations": ["Uses TypeScript"]}]}' \
npx -y @modelcontextprotocol/server-memory
Search (memory server):
mcp call search_nodes --params '{"query": "TypeScript"}' \
npx -y @modelcontextprotocol/server-memory
For nested objects and arrays, ensure valid JSON:
mcp call edit_file --params '{
"path": "/tmp/file.txt",
"edits": [
{"oldText": "foo", "newText": "bar"},
{"oldText": "baz", "newText": "qux"}
]
}' npx -y @modelcontextprotocol/server-filesystem /tmp
# Table (default, human-readable)
mcp call <tool> --params '{}' <server>
# JSON (for parsing)
mcp call <tool> --params '{}' -f json <server>
# Pretty JSON (readable JSON)
mcp call <tool> --params '{}' -f pretty <server>
# List available resources
mcp resources <server-command>
# Read a specific resource
mcp read-resource <resource-uri> <server-command>
# Alternative syntax
mcp call resource:<resource-uri> <server-command>
# List available prompts
mcp prompts <server-command>
# Get a prompt (may require arguments)
mcp get-prompt <prompt-name> <server-command>
# With parameters
mcp get-prompt <prompt-name> --params '{"arg": "value"}' <server-command>
If using a server frequently during a session:
# Create alias
mcp alias add fs npx -y @modelcontextprotocol/server-filesystem /home/user
# Use alias
mcp tools fs
mcp call read_file --params '{"path": "README.md"}' fs
# List aliases
mcp alias list
# Remove when done
mcp alias remove fs
Aliases are stored in ~/.mcpt/aliases.json.
mcp tools --auth-user "username:password" https://api.example.com/mcp
mcp tools --auth-header "Bearer your-token-here" https://api.example.com/mcp
mcp tools docker run -i --rm \
-e GITHUB_PERSONAL_ACCESS_TOKEN="$GITHUB_TOKEN" \
ghcr.io/github/github-mcp-server
mcp tools npx -y @modelcontextprotocol/server-filesystem /tmp
mcp tools https://example.com/mcp
mcp tools http://localhost:3001/sse
# Or explicitly:
mcp tools --transport sse http://localhost:3001
# Allow access to specific directory
mcp tools npx -y @modelcontextprotocol/server-filesystem /path/to/allow
mcp tools npx -y @modelcontextprotocol/server-memory
export GITHUB_PERSONAL_ACCESS_TOKEN="your-token"
mcp tools docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server
export BRAVE_API_KEY="your-key"
mcp tools npx -y @anthropic/mcp-server-brave-search
mcp tools npx -y @anthropic/mcp-server-puppeteer
Before calling tools, run mcp tools to understand what's available and the exact parameter schema.
When you need to process results programmatically:
mcp call <tool> --params '{}' -f json <server> | jq '.field'
The table output shows parameter signatures. Match them exactly:
param:str = stringparam:num = numberparam:bool = booleanparam:str[] = array of strings[param:str] = optional parameterTool calls may fail. Check exit codes and stderr:
if ! result=$(mcp call tool --params '{}' server 2>&1); then
echo "Error: $result"
fi
If making several calls to the same server:
mcp alias add tmp-server npx -y @modelcontextprotocol/server-filesystem /tmp
mcp call list_directory --params '{"path": "/tmp"}' tmp-server
mcp call read_file --params '{"path": "/tmp/file.txt"}' tmp-server
mcp alias remove tmp-server
For safety, limit what tools are accessible:
# Only allow read operations
mcp guard --allow 'tools:read_*,list_*' --deny 'tools:write_*,delete_*' \
npx -y @modelcontextprotocol/server-filesystem /home
mcp tools --server-logs <server-command>
cat ~/.mcpt/aliases.json
Use --format pretty for detailed JSON output to debug parameter issues.
| Action | Command |
|---|---|
| List tools | mcp tools <server> |
| List resources | mcp resources <server> |
| List prompts | mcp prompts <server> |
| Call tool | mcp call <tool> --params '<json>' <server> |
| Read resource | mcp read-resource <uri> <server> |
| Get prompt | mcp get-prompt <name> <server> |
| Add alias | mcp alias add <name> <server-command> |
| Remove alias | mcp alias remove <name> |
| JSON output | Add -f json or -f pretty |
# 1. Discover what's available
mcp tools npx -y @modelcontextprotocol/server-filesystem /home/user/project
# 2. Check for resources
mcp resources npx -y @modelcontextprotocol/server-filesystem /home/user/project
# 3. Create alias for convenience
mcp alias add proj npx -y @modelcontextprotocol/server-filesystem /home/user/project
# 4. Explore directory structure
mcp call directory_tree --params '{"path": "/home/user/project"}' proj
# 5. Read specific files
mcp call read_file --params '{"path": "/home/user/project/README.md"}' proj
# 6. Search for patterns
mcp call search_files --params '{"path": "/home/user/project", "pattern": "**/*.ts"}' proj
# 7. Clean up alias
mcp alias remove proj
Ensure PATH is set: export PATH="$HOME/.local/bin:$PATH"
--params "$(cat params.json)"Some servers take time to start. The mcp CLI waits for initialization automatically.
For filesystem server, ensure the allowed directory path is correct and accessible.