Teach coding agents when to use Decodo with Agent Skills

Last tested and updated: August 25, 2026

An MCP server gives an agent callable tools. An Agent Skill gives it reusable instructions for deciding when and how to use those tools. You can use either one alone, but they solve different problems.

Layer	What it provides	Where the secret belongs
Decodo MCP server	Named scraping and search tools inside an MCP client	MCP environment or client secret storage
Decodo CLI or raw API	Commands or HTTP requests for code and automation	Environment variable or external secret manager
Decodo Agent Skill	Routing rules, setup guidance, checks, and workflow procedure	Nowhere in the Skill

Complete the MCP connection lesson first if you want the Skill to route through MCP. A Skill cannot create a working credential or transport by itself.

Step 1: choose the Skill that matches the job

Decodo’s official repository currently contains two Skills:

decodo-web-scraping — the core routing layer for general pages, Google and Bing, ecommerce targets, Reddit, TikTok, and YouTube. It can route among the Decodo CLI, hosted MCP server, and raw HTTP API according to the agent’s environment.
decodo-price-monitoring — a pricing workflow for current prices, seller comparison, and recurring price checks. It builds on decodo-web-scraping.

Install the core Skill first. Add price monitoring only when you have a defined product-matching rule, schedule, storage destination, and alert threshold.

Step 2A: install the plugin in Claude Code

Decodo recommends its plugin path for Claude Code. Run these slash commands inside Claude Code:

/plugin marketplace add Decodo/agent-skills
/plugin install decodo@decodo-skills

The first command adds Decodo’s repository as a plugin marketplace. The second installs the decodo plugin from the decodo-skills marketplace. Restart or reload plugins if Claude Code asks.

Step 2B: install manually in another supported agent

Decodo describes the Skills as compatible with Claude Code, Cursor, Codex, Gemini CLI, Windsurf, Claude Desktop, and claude.ai. Support does not mean those products share one folder or installation command.

Use this portable process:

Open or clone the official Decodo/agent-skills repository.
Review the selected SKILL.md and its referenced files before installing it. A Skill is executable guidance for your agent, so treat it like code.
Copy the complete skills/decodo-web-scraping directory into the user- or project-level Skills directory recognized by your client.
Install decodo-price-monitoring beside it only if you need that workflow.
Restart the agent or reload Skills, then ask it to list or explain the installed Decodo Skill.

For example, Codex uses $HOME/.agents/skills/ for user-level Skills and .agents/skills/ for repository Skills, while Claude Code uses ~/.claude/skills/. Cursor, Windsurf, and other clients can change their discovery locations and interfaces, so verify their current Skills documentation instead of guessing a path.

Choose scope deliberately:

User level: available across your projects; useful on one trusted personal machine.
Project level: travels with one workflow and can be reviewed by teammates; never include the token.

Step 3: keep the credential outside the Skill

Installing a Skill does not authenticate Decodo. Keep the Web Scraping API Basic token in the environment, MCP client secret storage, or a dedicated secret manager.

The same authentication distinction still applies:

Local MCP uses the raw token as SCRAPER_API_TOKEN.
Hosted MCP sends Authorization: Basic <token>.
Raw API and CLI routes follow their own documented environment or request format.

Never paste a real token into SKILL.md, a project instructions file, a chat prompt, a screenshot, or a command that will remain in shell history. Rotate the token if it was exposed.

Step 4: test the routing decision

Start with one small, public target:

Use the installed Decodo web-scraping Skill to collect the first five visible
story titles and destination links from https://news.ycombinator.com/.
Before calling anything, state which Decodo surface you selected and why.
Use at most one Decodo request. Do not fall back to another web tool.
Return the tool or command used, source URL, observed time, and five results.

Accept the test only when:

The agent identifies a Decodo route available in its actual environment.
The execution log shows the expected Decodo tool, CLI command, or API request.
It returns five linked results without a silent fallback.
The request count stays inside the stated budget.
No credential appears in output or logs.

Instructions saying “use Decodo” are not proof that Decodo ran. Inspect the tool or command record.

Step 5: test the specialized workflow separately

For decodo-price-monitoring, begin with a single observation rather than a schedule:

Use the Decodo price-monitoring Skill to check Amazon product B09H74FXNW once.
Return the exact product identity, local price and currency, seller, stock state,
product URL, observed time, and selected Decodo route. Use at most two Decodo
requests. Do not create a schedule or alert yet.

Verify product identity and source evidence before adding recurring execution. Scheduling an unverified match only repeats the mistake automatically.

Step 6: update Skills like code

An installed Skill is a snapshot. When Decodo changes the upstream repository:

Read the release or commit diff.
Check whether routing, commands, credential names, or supported targets changed.
Update the installed directory.
Repeat the one-request smoke test.

Do not auto-update third-party instructions directly into a production workflow without review. A new routing rule can change cost, data handling, or which external service the agent calls.

When not to add a Skill

Skip the Agent Skill when:

you only need one explicit MCP tool call;
your application already implements deterministic API routing;
the client cannot discover Skills reliably;
you cannot inspect which route or request the agent used.

In those cases, a precise prompt or direct API call is easier to audit. Add a Skill when repeatable routing logic is the part you want the agent to remember.