withmcp

withmcp is a tool for managing your system-wide collection of MCP servers, allowing easy toggling of individual servers or groups of them, globally or per project.

❯ withmcp
Usage:
  withmcp [options] [--] <harness> [args...]

Configured Servers:
  excalidraw  https://api.excalidraw.com/api/v1/mcp  (not enabled)
  github      https://api.githubcopilot.com/mcp/  (not enabled)
  linear      https://mcp.linear.app/mcp  (not enabled)
  playwright  npx @playwright/mcp@latest  (not enabled)
  slack       https://mcp.slack.com/mcp  (not enabled)


❯ withmcp +github claude   # launch claude with linear MCP enabled

❯ withmcp +webdev codex    # launch codex with playwright and github

Why would you want to do that? These were my motivating examples:

• Found a cool server that has very specific use cases so you don't want it in the context by default? -> Save it in your config, enable when needed.

• Want to switch between different toolsets for different tasks? -> Save them in your config, enable toolsets when needed.

• Want to enable an MCP server globally like voice mode when working remotely? -> Save it in your config, enable when needed.

Quick start

cargo install --git https://github.com/lava/withmcp
withmcp edit  # create ~/.config/withmcp/profiles/default.toml

Create a minimal profile:

[servers.linear]
url = "https://mcp.linear.app/mcp"

Run with that server enabled:

withmcp +linear claude

Supported agents

Currently claude, codex and pi are supported.

Usage

Run a one-off session with a given server enabled:

❯ withmcp +linear claude

Permanently enable it:

❯ withmcp enable linear

All sessions launched by withmcp will now have it. The inverse also works:

❯ withmcp -linear claude   # run without linear mcp

You can keep separate profiles that define their own set of servers:

withmcp -p work codex

You can pull in servers from other profiles for a one-off run:

withmcp +work/slack claude

How it works

withmcp is a thin launcher for the selected agent harness. It generates a list of MCP servers that should be enabled for the current directory by looking at its own config. It then passes that configuration in the required harness-specific format to the agent.

Note that it does not do anything after launching the requested harness, so there is no way to update the config at runtime. For that, you would need to use a MCP proxy server.

It also does not attempt to perform any communication with the MCP servers themselves. In particular, authentication still happens completely inside your harness of choice, after launching.

When it detects that a given MCP server is already configured natively for the selected harness, it will be left out of the generated config.

Building

cargo install --path .
rustup target add x86_64-unknown-linux-musl
cargo build --release --target x86_64-unknown-linux-musl

Known limitations

• Codex receives servers as command-line arguments, so expanded ${VAR} and $(command) values are visible in the process list.
• Detection of servers the harness already defines is best-effort. For example, Claude Code plugins and managed configs are not checked.
• Pi uses a copy of its global mcp.json while withmcp runs, so changes the adapter writes to that file during the session (e.g. from /mcp setup) are lost. Servers pulled in through the file's imports are not checked for collisions.
• list does not check for collisions with a harness's own servers; use which <harness> for that.

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option.