boxxkite-mcp
An MCP server over a hosted boxxkite control-plane — lets any MCP-compatible client (Claude Code, Claude Desktop, Codex, Cursor, etc.) attach a real sandboxed code-execution backend as a native tool source, zero custom integration code.
Prefer no local install? A control-plane deployment built from this repo also exposes a remote Streamable HTTP MCP endpoint directly at
https://your-control-plane.example.com/mcp/— add that URL to your MCP client's config instead of installing this package. Seedocs/HOSTED-MCP-DESIGN.md. Use this package when you want the MCP server process running on your own machine instead.
Install
pip install boxxkite-mcp
# or, to run it as a standalone MCP server without a project venv:
pipx install boxxkite-mcp
Configuration
Two required environment variables:
| Variable | Meaning |
|---|---|
BOXXKITE_BASE_URL |
Base URL of the boxxkite control-plane |
BOXXKITE_API_KEY |
A bxk_live_... API key for your account |
Run
BOXXKITE_BASE_URL=https://your-control-plane.example.com \
BOXXKITE_API_KEY=bxk_live_... \
boxxkite-mcp
Speaks MCP over stdio — point an MCP client's config at the boxxkite-mcp command.
Tools
Sandbox lifecycle and exec/file tools — create_sandbox, destroy_sandbox,
get_sandbox, list_sandboxes, exec, file_create, view, str_replace,
ls, glob, grep — every per-sandbox tool takes session_id as a
parameter, so the calling agent owns the full lifecycle within one
conversation.
Custom image tools (build a sandbox image with extra packages baked in,
then pass its id as create_sandbox's image_id) — create_sandbox_image,
get_sandbox_image, list_sandbox_images, delete_sandbox_image.
Independent storage volume tools (create persistent storage mountable into
one or more sandboxes via create_sandbox's volume_mounts) —
create_sandbox_volume, get_sandbox_volume, list_sandbox_volumes,
delete_sandbox_volume.
Outbound-MCP connection tools (grant a sandbox network egress to a curated
MCP catalog entry via create_sandbox's mcp_connection_names — see
docs/OUTBOUND-MCP-DESIGN.md;
there is no MCP-proxy transport yet, so this only widens network reachability,
it doesn't yet let the sandbox speak MCP protocol to the destination) —
create_mcp_connection, list_mcp_connections, delete_mcp_connection.
Language-server (LSP) tools for code intelligence inside a sandbox — start a
language server, open a file into it, request completions at a position, then
stop it — lsp_start, lsp_open, lsp_completion, lsp_stop. Like the other
per-sandbox tools, each takes session_id.
MemoryBase tools — durable, account-scoped memory that survives past any one
sandbox or conversation (see
the developer guide for the
full model: opt-in, self-hosted retrieval by default, no third-party memory
API) — remember, recall, ingest_memory, memory_profile,
forget_memory. The server's MCP instructions steer a connecting agent to
call recall/memory_profile early in a task and remember when it learns
something durable, but nothing here changes what the account's own memory
retrieval settings do server-side.
That's 31 tools in total.
Security
exec runs arbitrary shell commands with no client-side allowlist — the
isolation boundary is the sandbox itself (see the root repo's SECURITY.md),
not these MCP tools' argument validation. exec/view results are returned
to the calling LLM as plain, unsanitized text — treat sandbox output as
untrusted input, the same as a web-fetch or file-read tool's result.
Related tools
Moving an in-progress local Claude Code/Codex CLI/opencode session (full
conversation history) into a fresh boxxkite sandbox is not something
this MCP server can do as a tool call: a handoff adapter needs to read
local, on-disk CLI session state (e.g. Claude Code's
~/.claude/projects/... files) on the user's own machine, while an MCP
tool call runs wherever the MCP client invokes it, and boxxkite-mcp itself
is a thin proxy to the hosted control-plane with no access to the calling
agent's local filesystem. That's handled instead by a local-only command in the main CLI,
boxxkite handoff <tool> — see
../docs/handoff-adapters.md for how it works.
Development
pip install -e ".[dev]"
pytest tests/
See the root README for what boxxkite is and the full self-hosting story.
Questions, bug reports, or need a usage-limit bump? Join the Discord.
Metadata
Release files for boxxkite-mcp 0.8.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| boxxkite_mcp-0.8.0.tar.gz | 20.5 kB | Without provenance | Details for boxxkite_mcp-0.8.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| boxxkite_mcp-0.8.0-py3-none-any.whl | Python 3 | none | any | Details for boxxkite_mcp-0.8.0-py3-none-any.whl |
Total release size: 33.9 kB
Release files / boxxkite_mcp-0.8.0.tar.gz
| Download URL | boxxkite_mcp-0.8.0.tar.gz |
|---|---|
| Size | 20.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
83c01e7b41840d8204e96ed27f09b6aec14d8f8a0bf47360b6a3e7a363a4dd55
|
|
BLAKE2b-256 checksum How to use checksums |
cba98572d4272c664859a211eab8c268bd580ac266852d1e46e076918158a492
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Release files / boxxkite_mcp-0.8.0-py3-none-any.whl
| Download URL | boxxkite_mcp-0.8.0-py3-none-any.whl |
|---|---|
| Size | 13.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
17b96ac82209616640f33ea33a4f57f17e4b320fe419c4c7d4a5b532e0d729c4
|
|
BLAKE2b-256 checksum How to use checksums |
9f13c9908212ba7812bfae0fc2a746d526f47636e6de987bc6f09fdfe4e7649e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.14.6
|
Security for boxxkite-mcp 0.8.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.