Skip to main content
Glama
DeepSeek Harness MCP Server

DeepSeek Harness MCP Server

dsh_submit_task

Destructive

Launch a bounded DeepSeek Harness coding task in an explicitly authorized workspace with relative scopes, constraints, timeout, and checks, then retrieve and independently verify results.

Instructions

Start one DeepSeek Harness coding task. You must explicitly specify the human-authorized absolute project/worktree in workspace on every call; it is this task's write directory, not the MCP server installation directory. Different tasks may specify different directories. Inspect the canonical workspace returned with task_id. Optional server roots further restrict admission. Supply relative allowed_paths and checks; get results with dsh_get_task and independently verify them. After an uncertain response, list tasks before retrying.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "workspace": "<workspace>",
      "objective": "<objective>",
      "allowed_paths": "<allowed_paths>"
    },
    "name": "dsh_submit_task"
  }
}

Input Schema

NameRequiredDescriptionDefault
objectiveYesOne small execution task with a clear expected result
workspaceYesRequired explicit authorization for this task: the absolute existing project/worktree directory the worker may modify. Never infer it from the MCP server cwd or reuse a previous task directory. Optional server roots remain an upper bound.
constraintsNo
allowed_pathsYesWorkspace-relative task scope; instructions to the worker, not an enforced file allowlist
timeout_secondsNo
reasoning_effortNoOptional exact Harness reasoning effort, e.g. off, low, high or max when advertised by the selected model. Omit to keep its default. Unsupported values fail before prompting; the confirmed effort is returned in task results.
acceptance_checksNoVerification instructions; never executed directly by this MCP server

Schema Changelog

Changes observed during successful MCP inspections.

  1. First observedv0.1.0

TDQS

A4.1/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The annotations already declare destructiveHint=true, openWorldHint=true, and idempotentHint=false, but the description adds real context beyond them: workspace is defined as the task's write directory rather than the server install directory, server roots act as an admission upper bound, allowed_paths is advisory rather than an enforced allowlist, and an ambiguous response should trigger a list-before-retry. What it does not state is the failure/timeout outcome or how partial writes are handled, which keeps it off a 5.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The purpose is front-loaded in the first sentence, which is good, but the remainder is a dense run-on chain of caveats packed with semicolons that an agent must parse carefully. Several clauses restate schema text (workspace authorization, relative allowed_paths) rather than adding new information, so some tightening is possible.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a destructive, open-world submission tool with no output schema, the description covers the critical ground: authorization, write scope, admission restrictions, retrieval via dsh_get_task, verification expectations, and safe retry after an ambiguous response. It leaves the failure/timeout aftermath and the point at which dsh_respond_permission becomes relevant unaddressed.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 71% and the description supplies meaning the schema cannot: workspace must be an explicit, human-authorized absolute path per call and must not be inferred or recycled, while allowed_paths is explicitly a worker instruction rather than an enforced file allowlist. It adds nothing about objective, constraints, or timeout_seconds, which are the undocumented remainder, so a 4 rather than a 5.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

The opening sentence states a specific verb and resource: "Start one DeepSeek Harness coding task." It also points to dsh_get_task for results, which separates this from the read-side sibling. It stops short of naming dsh_cancel_task or dsh_list_tasks as lifecycle alternatives, so differentiation is good but not exhaustive.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit preconditions (human-authorized absolute project/worktree required on every call), an explicit anti-pattern (never infer from the MCP server cwd or reuse a previous task directory), and an explicit recovery path ("After an uncertain response, list tasks before retrying"). It also routes result retrieval to dsh_get_task and tells the caller to independently verify, which is unusually concrete guidance for a submission tool.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.