dsh-multica-runtime: A Multica Runtime Bridge Plugin for DeepSeek Harness
multica-ai/dsh-multica-runtime
DSH plugin bridging Multica and DeepSeek Harness via a private stdio runtime.
Enables Multica to securely invoke DeepSeek Harness capabilities over stdio without modifying DSH source code.
dsh plugin --profile web add github:multica-ai/dsh-multica-runtime- Category
- Integrations & Connections
- Platform
- DSH-Plugin
- Author
- multica-ai
- Distribution
- Plugin
dsh-multica-runtime Key Features
dsh-multica-runtime Repository

dsh-multica-runtime Repo Summary
dsh-multica-runtime is an integration DSH plugin for DeepSeek Harness (DSH), maintained by multica-ai and written in TypeScript. It provides a private, out-of-tree runtime bridge between Multica and DeepSeek Harness, solving the problem of allowing Multica to invoke DSH capabilities over stdio without modifying the DeepSeek Harness source code. Built on @deepseek-ai/dsh-base, it exposes a versioned JSONL protocol and supports model and thinking-level discovery, committed text, reasoning, tool, result, and token-usage events, as well as cooperative cancellation and durable session resume. It also translates Multica's MCP configuration into DSH stdio or streamable-HTTP clients, and forwards only Multica's server-minted mat_ task token into DSH's credential-scrubbed shell, preserving task attribution without exposing model-provider credentials.
What core features does the dsh-multica-runtime DSH plugin support?
The plugin supports interaction with DSH via --probe, --list-models, and --stdio commands, where --probe must return protocol version 1 for Multica to discover the profile. The runtime contract includes: model and thinking-level discovery from DSH itself; committed text, reasoning, tool, result, and token-usage events; cooperative cancellation and durable session resume; canonical Multica MCP configuration translated to DSH stdio or streamable-HTTP clients; per-runtime/agent session roots supplied by the Multica daemon; headless one-shot approvals with no interactive question surface. Additionally, it narrowly forwards only Multica's server-minted mat_ task token into DSH's otherwise credential-scrubbed shell, so in-task multica commands retain task attribution without exposing model-provider credentials.
How to install and configure dsh-multica-runtime?
To install this DSH plugin, you need to build the local package and then add it to a DSH profile using the DSH command line. The commands are as follows:
pnpm install
pnpm check
pnpm build
dsh plugin --profile multica add /absolute/path/to/multica-dsh-runtime
For non-standard DSH installations, you can point the daemon at its launcher using the environment variable MULTICA_DSH_PATH. The plugin is currently validated against @deepseek-ai/[email protected] and its matching @deepseek-ai/dsh-* package family.
How does dsh-multica-runtime ensure privacy and security?
This repository contains only the Multica integration layer and does not vendor or redistribute DeepSeek Harness source code. It disables DSH telemetry via the bundle patch, and stdout is protocol-only while diagnostics go to stderr. It strictly forbids committing API keys, MCP secrets, session logs, or generated profiles. DEEPSEEK_API_KEY is read by DSH's credential provider at process runtime and must not be stored in this repository.
What is the development and debugging workflow for dsh-multica-runtime?
For local development, you can use pnpm install, pnpm check, and pnpm build to build and check. After building, you can debug using dsh --profile multica --probe, dsh --profile multica --list-models, and dsh --profile multica --stdio. The local .local/ directory is ignored and may hold an isolated DSH home and a development launcher, but neither belongs in source control.
What is the compatibility and maintenance status of dsh-multica-runtime?
This plugin is maintained by multica-ai, written in TypeScript, and last updated on 2026-08-14, with currently 1 open issue. It depends on multiple official DSH packages such as @deepseek-ai/dsh-agent, @deepseek-ai/dsh-cmdline, @deepseek-ai/dsh-llm, @deepseek-ai/dsh-mcp-client, etc., ensuring compatibility with the DeepSeek Harness ecosystem.