Dynamic MCP Power

This power teaches an agent how to leverage a registry of MCP server tools without the overhead of having to pre-start servers and pre-load all tool descriptions into model context. It can also be used with MCP “code mode” where an AI agent can write programs that stich together local MCP tool calls.

Use the mcp-cli CLI to access a list of configured MCP servers.

Use the mcp tools CLI to dynamically start and invoke tools.

Onboarding

To use this power, ensure you have the following prerequisites installed:

• mcp-cli – you can install the binary (written in go) from the GitHub releases page.

• mcp tools cli – You can install it using HomeBrew

brew tap f/mcptools
brew install mcp

Discovery

Use the mcp-cli tool to look up available mcp servers using the mcp-cli ls -acd command.

# List all available servers and commands to start them
mcp-cli ls -acd

NAME                    COMMAND                                                                               DESCRIPTION
----                    -------                                                                               -----------
time                    uvx mcp-server-time --local-timezone America/New_York                                 Provides current time and timezone utilities.
aws-agentcore           FASTMCP_LOG_LEVEL=ERROR uvx awslabs.amazon-bedrock-agentcore-mcp-server@latest        Amazon Bedrock AgentCore platform documentation and management

Invocation

Invoke MCP server tools via the mcp CLI.

One-Shot Calls

# List available tools from servers using the server commands exposed by `mcp-cli ls -acd`
mcp tools <server-command>

# Call a tool
mcp call <tool-name> -p '<json-args>' <server-command>

# With longer timeout for slow servers (default 30s)
mcp call -t 60 <tool-name> -p '<json-args>' <server-command>

Environment Variables

The mcp tools command does not currently support inline envvars (when not using containers), so you’ll need to export them first. For example, instead of:

mcp tools AWS_REGION=us-east-1 FASTMCP_LOG_LEVEL=ERROR uvx awslabs.aws-api-mcp-server@latest

You’ll need to run:

export AWS_REGION=us-east-1; export FASTMCP_LOG_LEVEL=ERROR; mcp tools uvx awslabs.aws-api-mcp-server@latest

For container-based servers, you can pass them inline. For example:

mcp tools docker run -i --rm -e AWS_REGION=us-east-1 -e FASTMCP_LOG_LEVEL=ERROR -v $HOME/.aws:/app/.aws public.ecr.aws/awslabs-mcp/awslabs/aws-api-mcp-server

Examples

mcp tools npx -y @modelcontextprotocol/server-filesystem /tmp
mcp call read_file -p '{"path": "/tmp/test.txt"}' npx -y @modelcontextprotocol/server-filesystem /tmp
mcp tools uvx mcp-server-git
mcp tools docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN=xyz ghcr.io/github/github-mcp-server

Clean Output

Disable ANSI color codes with NO_COLOR=1:

NO_COLOR=1 mcp call InternalSearch -p '{"query": "test"}' aim mcp start-server builder-mcp