dsh-mcp-bridge
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@dsh-mcp-bridgesearch memory for the API rate limit decision"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
dsh-mcp-bridge
Hands a DeepSeek Harness deployment's capabilities to any MCP client — memory, the durable task ledger, project-agent dispatch and the wider MCP estate — as declared tools instead of shell commands a model has to be told about and remember.
Why
claude -p runs its own agent loop, so it never receives the harness's tool schemas: a
Claude-backed session has no memory, no ledger, no dispatch. It does accept MCP servers.
This is that server.
Everything it exposes already has a command-line entry point, so this adds a surface, not authority — it can do exactly what the caller could already do in a shell. What it cannot bridge is anything that only exists inside a live harness turn (goals, AgentTeams); those are session-scoped by design and have no external surface.
Related MCP server: WEATHGARDS
Tools
tool | does |
| the durable memory store |
| the task ledger |
| hand work to a project agent, then poll it |
| any server in the MCP estate, uncomposed |
| the repository ownership registry |
| reach the user |
Use
// mcp-bridge.json
{ "mcpServers": { "dsh": { "command": "node", "args": ["/path/to/src/server.js"] } } }claude -p --strict-mcp-config --mcp-config ./mcp-bridge.jsonPoint DSH_JARVIS_BIN at the directory holding the CLIs if it is not ~/.dsh-jarvis/bin.
Three things learned the hard way
--mcp-configis variadic. Inline JSON followed by a positional prompt makes the client read the prompt as a second config path (MCP config file not found: /tmp/List the tool…). Pass a file.Don't exit the instant stdin reports EOF. The spec says client-closes-stdin is shutdown, but some clients hand the server a stdin already at EOF; the server exits before the first request, the client respawns it, and it loops. Observed: 32 spawns in three minutes, no call completed. Here EOF only ends the process once something has been served, with a bounded idle timeout so an abandoned server still goes away.
Verify a bridge with a write, never a read. A model asked to "call the search tool" can reproduce plausible output from its own context when the call actually failed — one did exactly that here and later admitted it. A write with a random token either lands in the store or does not.
Tests
npm test20 tests, no dependencies, no client required: the tool catalogue and the JSON-RPC layer are both pure.
License
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Your org's AI agents, tasks, runs, search, and brain files as MCP tools and resources.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Agent-native notes, tasks, dev-docs, vaults, sync & handoffs. MCP + OpenAPI dual surface.
MCP delegation fallback for AI agents to discover capabilities, knowledge, tools, and collaborators.
Related MCP Servers
- AlicenseBqualityFmaintenanceExposes OpenClaw tools (web search, page fetching, browser automation, PDF analysis, messaging, inter-agent communication, Canvas) as MCP tools for ACP agents.7-
- FlicenseNot gradedqualityDmaintenanceExposes MCP tools that enable remote LLMs to query local Docker containers, OS processes, and system services in real time.-
- AlicenseAqualityDmaintenanceExposes Agent Skills to AI agents as MCP tools, enabling discovery and activation of skill instructions for coding agents.327MIT
- FlicenseNot gradedqualityCmaintenanceExposes task management (add, list, complete tasks) and document search (RAG) as MCP tools for AI agents.-