Zukko Blog
Guides to Claude Code, Cursor, Codex and the Claude and GPT APIs: setup, token savings and troubleshooting.
Cline "hit repeated tool call failures": causes and fixes
6 min · · errors
The message means the agent failed several tool calls in a row. Causes and a checking order, from the base URL to context size.
Langflow or LangChain: a visual builder versus code
6 min · · сравнение
Langflow builds an app on a canvas, LangChain is written in code. Who each approach suits, and how to connect the gateway in both.
Qwen Code: error 401 invalid access token or token expired
5 min · · errors
A 401 in Qwen Code often means a stale Qwen OAuth login or a key sent to the wrong address. Step-by-step diagnosis and a switch to your own key.
Unable to connect to API (ConnectionRefused) in Claude: causes and fixes
6 min · · errors
Claude Code cannot open a connection to the API. We check the address, proxy, DNS and network with three curl commands and find the cause from the result.
Dify in Docker: installation, first-run setup and connecting models
6 min · · development
Install Dify with Docker Compose, create the admin account and connect Claude, GPT and DeepSeek models on one key. Plus what to do when containers will not start.
Project rules in Qwen Code: context files and token savings
5 min · · агенты
How to explain project rules to Qwen Code once: QWEN.md, path-conditional rules, skills and MCP, and how many tokens each such file costs.
Refactoring code with AI: small steps and tests
5 min · · development
An agent can rewrite a module in a minute and break it just as fast. How to organize refactoring: tests first, one kind of change at a time, a commit per step, rollback.
AI for code analysis: how to understand someone else's project faster
5 min · · development
How to get a map of an unfamiliar codebase in an evening: prompts for modules and entry points, checking answers against file references, and cost with and without cache.
AI for fixing code: how to hand over an error and get a working fix
5 min · · development
Fix quality depends on what you show the model. The data to include about the error, a prompt template, asking for a minimal diff, and verifying with a test.
MCP servers for Codex CLI: config.toml setup step by step
5 min · · development
Codex connects MCP servers through the codex mcp add command or tables in config.toml. Both ways, how to verify, and how many tokens tool descriptions use.
opencode or Codex CLI: which terminal agent to choose
6 min · · агенты
Both agents are open source, both read AGENTS.md and both work with your own API. The differences are in providers, modes and cache billing. We compare and price a session.
Cursor CLI: the Cursor agent in the terminal, install and setup
5 min · · агенты
The Cursor agent works in the terminal as well as the editor, with the agent command. Install it, configure it, run it in scripts and compare it with Claude Code and Codex.
GLM or DeepSeek via API: comparing price, context and strength index
5 min · · models
GLM charges the same for input and output, DeepSeek is cheaper on input. Two worked scenarios show which wins where and when a 128K window is not enough.
Claude Sonnet 4.6: what it is and when to pick it over Sonnet 5 and 5.5
6 min · · models
Sonnet 4.6 is still in the Zukko catalog, but it costs more and scores lower than Sonnet 5 and 5.5. When to leave a working pipeline alone and when to move.
Cline in Russia: installation and model setup without a foreign card
6 min · · IDE
The Cline extension runs locally, so the only question is where model requests go and how you pay for them. Step-by-step setup with one key and a task cost estimate.
Regular expressions with AI: how to get a working regex and test it
5 min · · development
A model writes regex fast but errs confidently on edge cases. A request with examples, an auto-check loop and protection against hangs.
Analyzing logs with AI: from thousands of lines to a clear cause of failure
4 min · · автоматизация
Do not send gigabytes of logs to a model. Filter, group by templates, ask for hypotheses tied to lines and verify them. Code and price estimate inside.
Business correspondence with AI: letters, replies and tone without bureaucratese
4 min · · автоматизация
How to get letters from a model that do not sound like boilerplate: request structure, ready prompts, a company tone profile and a pre-send check.
AI for DevOps: configs, scripts and incident analysis
5 min · · автоматизация
A model drafts configs and reads logs well, but must not hold production rights. Tasks, safe boundaries and a sample script.
Cline or Continue: agent or assistant, which to pick in VS Code
6 min · · сравнение
Cline takes a whole task and asks for permission, Continue offers chat, edit and autocomplete. How models, config and token spend differ between them.
LangChain.js: connect to an OpenAI-compatible API in Node.js
4 min · · development
How to point ChatOpenAI from LangChain.js at the Zukko gateway: one baseURL setting, streaming, tool calls and a spend check.
An MCP server for Bitrix24: CRM deals and tasks in an agent's hands
6 min · · агенты
The official Bitrix24 MCP gives an agent documentation, not portal data. How to write your own server for deals and tasks and keep the agent on a short leash.
opencode or Cursor: a console agent or an editor with AI
5 min · · сравнение
One lives in the terminal and works with any provider, the other brings a whole editor. A comparison by scenarios, model setup and price.
Function calling or MCP: what is the difference and what to choose
5 min · · агенты
These are not two rival methods but two levels of the same task. Who describes a tool, who runs it, and when you need MCP.