Cicall
A terminal-independent MCP runtime for CLI agents, enabling orchestration of CLI processes via a background daemon and any MCP client.
README
Cicall
Run and orchestrate CLI agents from any terminal.
Cicall is a terminal-independent, local MCP runtime for CLI agents. It owns the agent processes and their pseudo-terminals in a small background daemon. Your terminal, editor, desktop app, or MCP client is only a replaceable frontend.
Claude Code / Codex / Kimi / OpenCode
│ MCP (stdio)
cicall-mcp
│ local Unix socket
cicall daemon
│ PTY
any CLI agent or command
This repository is an independent extraction of the session-orchestration idea first prototyped in Helios. Cicall does not depend on Helios, Ghostty, AppKit, or any particular terminal emulator.
Status
Cicall is an early, usable MVP for macOS and Linux. The current release is local-only and has no network listener. Do not expose its Unix socket to users you do not trust: a connected client can launch commands with your account.
Install
Python 3.9 or newer is required. The simplest install keeps Cicall in its own
environment and exposes the commands through ~/.local/bin:
curl -fsSL https://raw.githubusercontent.com/basionwang-bot/cicall/main/scripts/install.sh | sh
~/.local/bin/cicall setup --client hermes --project my-app --cwd "$PWD"
You can inspect scripts/install.sh before running it. A manual installation from the GitHub release is documented in docs/TESTING.zh-CN.md.
cicall setup starts the local runtime and prints the exact MCP configuration
for the current installation. The MCP server also starts the daemon automatically,
so users do not have to manage it before opening their MCP client.
For development from a source checkout:
cd ~/Desktop/cicall
python3 -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
pip install -e .
cicall setup --client generic --project my-app
No Python package is required at runtime beyond the standard library.
Use it from any terminal
Start an interactive CLI and stay attached:
cicall run --project my-app --cwd ~/code/my-app -- claude
Start it in the background:
cicall run --detach --project my-app --cwd ~/code/my-app -- codex
cicall list --project my-app
cicall attach s-xxxxxxxxxxxx
Read or drive a session without attaching a terminal UI:
cicall read s-xxxxxxxxxxxx
cicall send s-xxxxxxxxxxxx "Please run the tests" --enter
cicall key s-xxxxxxxxxxxx ctrl-c
cicall close s-xxxxxxxxxxxx
Switching from Terminal.app to iTerm, Ghostty, Warp, VS Code, or SSH does not
change the managed process. Run cicall attach SESSION_ID from the new terminal.
Connect an MCP client
Running cicall setup prints a ready-to-paste configuration. A manual example:
cicall setup --client hermes --project my-app --cwd "$PWD"
Supported setup targets are generic, hermes, claude, and codex. Codex
can be registered automatically:
cicall setup --client codex --project my-app --cwd "$PWD" --apply
Check an installation at any time:
cicall doctor --fix
Then register the server with a fixed project scope:
{
"mcpServers": {
"cicall": {
"command": "/absolute/path/to/cicall-mcp",
"args": ["--project", "my-app", "--cwd", "/absolute/path/to/my-app"]
}
}
}
The MCP server exposes:
list_sessionsread_outputsend_textsend_keystart_sessionclose_session
Cicall follows the standard newline-delimited MCP stdio transport. It also
auto-detects LSP-style Content-Length framing for compatibility with hosts
such as HermesPet; replies use the same framing selected by the client.
An MCP server started with --project my-app cannot read, control, or close
sessions belonging to another project.
With --cwd configured, an agent can start a CLI with only its argv:
{"command": ["claude"]}
Configuration
| Variable | Meaning | Default |
|---|---|---|
CICALL_HOME |
Runtime state directory | ~/.local/state/cicall |
CICALL_SOCKET |
Unix socket path | $CICALL_HOME/cicall.sock |
CICALL_PROJECT |
Default MCP project scope | current directory name |
The daemon passes CICALL_SOCKET, CICALL_PROJECT, and CICALL_SESSION_ID to
every managed child. An MCP client launched inside a managed session therefore
inherits the correct runtime and project automatically.
Architecture and boundaries
Cicall deliberately separates three concerns:
daemon.pyowns PTYs, process lifetime, output buffers, and the local socket.mcp.pytranslates MCP tool calls into runtime operations.cli.pyis one human-facing client; it can be replaced by a GUI or web UI.
The PTY is an execution backend, not a terminal emulator. Cicall does not render terminal output. This is what makes it independent from the terminal application.
Roadmap
- Persist and restore session metadata across daemon restarts.
- Add task/delegation tools above the session primitives.
- Add structured activity detection and
wait_until_idle. - Add Docker, SSH, tmux, and remote-agent backends behind a backend interface.
- Add HTTP/SSE MCP transport with authentication for remote deployments.
- Ship signed binaries, Homebrew packaging, and Linux packages.
- Add a protocol SDK so third-party GUIs can become Cicall frontends.
Contributing
Issues and pull requests are welcome. Keep the local runtime dependency-light, make dangerous capabilities explicit, and preserve project isolation.
License
MIT License. See LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.