# Shrimp Task Manager > Turn a vague goal into a dependency-tracked task graph an agent can execute step by step. [Canonical HTML page](https://top-mcps.com/mcp/shrimp-task-manager) · [server.json](https://top-mcps.com/mcp/shrimp-task-manager.json) · [methodology](https://top-mcps.com/about/methodology) ## Install ### Claude Desktop — `claude_desktop_config.json` Paste under mcpServers. Fully quit and reopen Claude after editing. ```json { "mcpServers": { "shrimp-task-manager": { "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### Claude Code — `CLI or .mcp.json` Run from your repo. Commit .mcp.json to share with your team. ```shell # export DATA_DIR=./shrimp-data claude mcp add shrimp-task-manager -- git clone https://github.com/cjo4m06/mcp-shrimp-task-manager && npm install && npm run build ``` ### Cursor — `.cursor/mcp.json` Global path: ~/.cursor/mcp.json. Reload window after editing. ```json { "mcpServers": { "shrimp-task-manager": { "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### VS Code — `.vscode/mcp.json` VS Code uses the "servers" key (not "mcpServers"). ```jsonc { "servers": { "shrimp-task-manager": { "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### Windsurf — `~/.codeium/windsurf/mcp_config.json` Open via Cascade → hammer icon → Configure. ```json { "mcpServers": { "shrimp-task-manager": { "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### Cline — `cline_mcp_settings.json` Open via the Cline sidebar → MCP Servers → Edit. ```json { "mcpServers": { "shrimp-task-manager": { "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### Continue — `~/.continue/config.json` Continue uses modelContextProtocolServers with a transport block. ```json { "experimental": { "modelContextProtocolServers": [ { "transport": { "type": "stdio", "command": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ], "env": { "DATA_DIR": "${DATA_DIR}" } } } ] } } ``` ### Codex CLI — `~/.codex/config.toml` Codex uses TOML. Each server is a [mcp_servers.] subtable. ```shell # ~/.codex/config.toml [mcp_servers.shrimp-task-manager] command = "git" args = [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build", ] env = { DATA_DIR = "${DATA_DIR}" } ``` ### Zed — `~/.config/zed/settings.json` Zed calls them "context_servers". Settings live-reload on save. ```jsonc { "context_servers": { "shrimp-task-manager": { "command": { "path": "git", "args": [ "clone", "https://github.com/cjo4m06/mcp-shrimp-task-manager", "&&", "npm", "install", "&&", "npm", "run", "build" ] }, "env": { "DATA_DIR": "${DATA_DIR}" } } } } ``` ### ChatGPT — `ChatGPT → Apps directory` Shrimp Task Manager doesn't ship a hosted HTTPS endpoint today. ChatGPT supports remote MCP servers only — to use this server in ChatGPT you'll need to deploy it to a public HTTPS URL first (e.g. via Cloudflare Workers or Vercel) or wait for an official remote build. ```none ``` ## At a glance - **Maintainer:** cjo4m06 (community) - **Transport:** stdio - **Auth model:** None - **Required secrets:** DATA_DIR - **Supported clients:** Claude, Claude Code, Cursor, VS Code, Any MCP-compatible client - **License:** MIT - **Language:** TypeScript - **Latest version:** latest - **Last verified:** 2026-06-22 - **GitHub stars:** 2,120 (fetched 2026-06-22T09:44:47.844Z) - **Score:** 62/100 (rubric 2026-04 — see https://top-mcps.com/about/methodology) - **Source:** https://github.com/cjo4m06/mcp-shrimp-task-manager ## Tools & permissions | Tool | Description | Args | Side effects | |------|-------------|------|--------------| | `plan_task` | Plan and analyze a goal before decomposition. | `description: string` | read | | `split_tasks` | Decompose a plan into dependency-tracked tasks. | `tasks: array` | write | | `execute_task` | Execute the next available task in dependency order. | `taskId: string` | write | | `verify_task` | Verify a task meets its completion criteria. | `taskId: string` | read | | `list_tasks` | List tasks with status and dependencies. | `—` | read | ## Security & scope - **Access scope:** read-write - **Sandbox:** Reads and writes the task graph to a local DATA_DIR. No outbound network calls of its own; the agent does the work, Shrimp tracks it. - **Gotchas:** - No npx package — the clone-and-build step is required, which makes it harder to pin to a version than registry-published servers. - The task graph lives in DATA_DIR as plain files; back it up if a long-running plan matters. - Single-author community project — evaluate maintenance before depending on it for production workflows. ## Quick answer **What it does.** Exposes tools to plan and analyze a goal, decompose it into tasks with dependencies, execute tasks in order, verify completion, and persist the whole task graph across sessions. Includes a research mode for systematic exploration before planning. **Best for:** - Decomposing vague specs into dependency-tracked tasks - Plan-then-execute coding loops - Cross-session continuity on multi-step builds - Research-mode exploration before planning - Auditable per-task verification **Not for:** - Single-shot tasks - Team collaboration workflows - Setups that cannot run a local clone-and-build step ## Description Shrimp Task Manager is an MCP server that converts natural-language goals into structured, dependency-tracked development tasks and then drives an agent through planning, execution, and verification. It is opinionated about chain-of-thought: the agent must plan and analyze before it decomposes, decompose before it executes, and verify each task on completion. Tasks persist across sessions with explicit dependencies, so a coding agent can pick up a partially-finished plan and continue without rediscovering what is left. ## Why it matters Agents are good at writing code and bad at sequencing work. Shrimp forces a plan-decompose-execute-verify loop and tracks dependencies between tasks, which is what stops an agent from skipping steps or redoing finished work on a multi-step build. ## Key features - Chain-of-thought planning workflow - Dependency tracking between tasks - Task decomposition into subtasks - Per-task completion verification - Persistent task graph across sessions - Research mode for systematic exploration ## FAQ ### How is it different from Claude Task Master? Both decompose work and persist state. Shrimp is more opinionated about the chain-of-thought workflow — it enforces a plan → analyze → decompose → execute → verify sequence and tracks explicit dependencies between tasks, plus a research mode before planning. ### Is there an npx package? No. As of mid-2026 there is no published npm package — you clone the repo, run npm install, and npm run build, then point your client at the built server. Budget a few extra minutes for setup. ### Where is the task graph stored? On disk in the configured DATA_DIR. The graph and its task memory persist across sessions, so the agent can resume a partial plan. ## Changelog - **2026-06-22** — Refreshed install snippets and fact sheet; verified for 2026. - **2025-03-15** — Initial directory listing.