KrakenMCP
Integrates with OpenAI Codex as a headless coding agent, allowing delegated tasks to run on GPT models; Codex support is listed as untested.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
KrakenMCP
Get more out of every AI subscription you pay for.
KrakenMCP lets your coding agent hand work off to other headless coding agents.
What: an MCP server that lets your agent delegate tasks to the other coding agents on your machine.
Why: save usage and context: spread the work efficiently across all your agent subscriptions, outside your chat.
How: your agent coordinates the session over MCP, and Kraken runs the other agents in headless mode for each task.
You keep talking to one agent, the "head", for example OpenCode running DeepSeek Flash. When a task needs planning, coding, a review or research, the head gives it to a "tentacle": another agent CLI that Kraken starts with the model you chose for that job, for example Claude Code running Opus. A strong model can plan and review while a cheap one writes the code. Only short summaries come back, so your chat stays small. Code is written on its own git branch, and nothing reaches your checkout until you approve the merge.
you <-> head: OpenCode + DeepSeek Flash
| MCP
v
KrakenMCP --> tentacle runs headless on, for example
think, plan Claude Code + Opus (read-only)
review Codex + GPT (read-only)
execute OpenCode + a cheap model (own git branch)
research Claude Code + Sonnet (web, cites sources)
your own Cursor, Aider, ...The head
The head is any coding agent on your machine that has Kraken connected as an MCP server: OpenCode, Claude Code, Codex, Cursor or any other MCP client. It's the one you chat with; it decides what to hand off and brings the results back to you. Use a cheap, fast model here and let the tentacles do the heavy work. To connect one, see Install.
Related MCP server: Roundtable AI MCP Server
Tentacles
Out of the box:
Tentacle | Does | Runs on (default) |
| Second opinion on a hard problem: options and a recommendation | Claude Code + Opus |
| Implementation plan, file by file | Claude Code + Opus |
| Writes the code on its own git branch | OpenCode |
| Reviews the change: approve or request changes | Claude Code + Opus |
| One-page map of part of the codebase | OpenCode |
| Searches the web, cites sources | Claude Code + Sonnet |
Add your own with krakenmcp edit <name>: it creates ~/.kraken/tentacles/<name>.md, a short
settings header and a prompt, and the head can use it right away:
---
description: Audit dependencies for known problems
access: read
agent: claude
model: sonnet
---
You are the AUDITOR. Check the dependencies of this repository: {{task}}
List each problem as "[severity] package - problem - fix".Supported agents: Claude Code, OpenCode, Codex (untested), Cursor CLI (untested), Aider (untested).
Install
You need Node.js 22.13+, git, and at least one supported agent CLI, installed and logged in.
npm install -g github:gateniomer/kraken
krakenmcp doctor # optional: check which agents Kraken findsThis installs straight from GitHub and adds the krakenmcp command. To update, run the same
install again.
Then connect it to your head. For OpenCode, add this to ~/.config/opencode/opencode.json:
{ "mcp": { "kraken": { "type": "local", "command": ["krakenmcp"], "enabled": true } } }For Claude Code: claude mcp add kraken -- krakenmcp. Other clients:
docs/reference.md.
Use it
Just ask the head. It knows when and how to use Kraken:
you: Add a dark mode toggle to the settings page.
head: → plan("add a dark mode toggle") runs on Claude Code + Opus
← a plan: theme setting, toggle, CSS variables
→ execute(plan) runs on OpenCode, on branch kraken/execute_7f3a
← done, tests pass
→ review(run) runs on Claude Code + Opus
← approve
head: Done and reviewed: 3 files changed. Merge it?
you: Yes.
head: → merge(run)
head: Merged into main.→ is a tool call the head makes, ← is what comes back.
To read what an agent wrote in full:
krakenmcp show # the newest result (a plan, a review, research)
krakenmcp show <run_id> --log # the agent's raw outputConfigure tentacles
Your tentacle files live in ~/.kraken/tentacles/.
krakenmcp list # every tentacle, its settings, and where each value came from
krakenmcp edit <name> # open your file for it (created with the current values), then check it
krakenmcp remove <name> # delete your file; a built-in goes back to its defaultsChange a built-in: write only what you want to change. For example
~/.kraken/tentacles/execute.md, to run your tests after every code change:
---
verify: npm test
---All fields are in docs/reference.md.
Other settings
krakenmcp mode auto # merge by itself when the review approves and verify passes (default: ask)
krakenmcp limit 3 # how many runs may work at once
krakenmcp prune # delete old logs and resultsSecurity
Code-writing agents auto-approve their own edits inside their worktree, verify runs shell
commands, and merge changes your checkout. Read SECURITY.md before using Kraken
on code you care about.
More
docs/reference.md: every tool, field, CLI command and agent adapter
CONTRIBUTING.md: code layout, tests, how to add an agent
npm install && npm test # development: unit + end-to-end tests (no tokens, about 2 minutes)Runs on macOS and Linux. Runs, messages and results live in a local SQLite database in
~/.kraken, so work survives restarts and Kraken itself sends nothing anywhere.
License
This server cannot be deployed
Maintenance
Related MCP Connectors
Coordination hub for AI coding agents: message teammates, ask humans, audit every event.
One shared project for a team's coding agents: decisions, a task board and handoffs.
Hand tasks, bugs and finished work to AI coding agents, and get back a write-up with evidence.
AI work orchestration for plans, tasks, teams, and coding-agent dispatch.
Related MCP Servers
- AlicenseAqualityBmaintenanceEnables AI assistants to delegate specific tasks to specialized sub-agents (e.g., test-writer, code-reviewer). Supports both Cursor and Claude Code with custom agent definitions.1315 npm98MIT
- FlicenseNot gradedqualityAmaintenanceEnables a primary coding assistant to delegate tasks in parallel to specialized local sub-agents, sharing project context and synthesizing their responses into a single result.-
- FlicenseNot gradedqualityAmaintenanceEnables a primary AI assistant to delegate tasks to specialized local AI sub-agents in parallel, share project context, and synthesize their responses from within an IDE.-
- FlicenseNot gradedqualityAmaintenanceEnables a primary AI assistant to delegate tasks to multiple specialized coding sub-agents, run them in parallel with shared project context, and synthesize their responses from within the IDE.-