Choose theme
AI Guide
How to add skills and tools to your AI assistant
Perhaps you want the same report format every week, or an assistant that can find a file without another copy and paste. Those are different needs. A skill can carry a working method; a tool connection can make information or actions available. This Guide follows one small task so you can see which part you actually need.
These are suggestions to adapt to your situation. Check important details against current original sources before acting.
Task
Choose an addition for one job
Start with the work that is missing: a repeatable method, access to information, or both.
Approach
Separate the method from the connection
A skill packages guidance and supporting material. MCP is a way for an app to connect to capabilities. They can work together.
Takeaway
Keep a comparison you can revisit
Use a small example, inspect the result and available activity, and record what the addition earned.
Start here
Choose a task you can judge without an add-on
This is for someone who can already ask an assistant a question and is wondering what skills or connected tools change. You do not need to install anything to explore the comparison below. For an actual trial, the chosen app must support the addition and its required setup.
Our example is a short workshop brief. You know the source notes, so you can notice an invented room number or a lost action. That makes the exercise easier to judge than an open-ended request to organize a whole project.
These fictional notes are deliberately incomplete. They name a day and time without a calendar date, and they do not confirm a room.
workshop-notes.txt
Workshop: Saturday, 10am
Mira: bring cards
Room: not confirmedIn this guide
A prompt, a skill and a connection do different jobs
A prompt gives the assistant the request in front of it. If you paste these notes and ask for a brief, the task can already be possible without an installation. A saved prompt may be enough when the work is occasional.
In the Agent Skills format, a skill is a folder with a SKILL.md file describing when it applies and how to do the work. It may also contain references, templates or executable scripts. The documented loading approach starts with a short description and brings in more instructions when the skill is relevant. Actual support and activation depend on the application.
MCP stands for Model Context Protocol. It gives an AI application a standard way to connect with a server that exposes capabilities. The documentation distinguishes callable tools, information resources and prompt templates. A file-reading operation is one possible capability; other tools can change files or act on services.
These are complementary parts, rather than a ladder from basic to advanced. A skill might explain how to prepare the brief while a connected tool retrieves its source. A skill can also use tools the app already has, without MCP. Adding instructions does not by itself configure a missing connection or grant account access.
In this guide
See what changes when you add each part
Compare the same workshop task with instructions and file access added separately or together. Notice where the notes come from, where the format comes from, and what still cannot be answered.
The illustrated drafts are written examples, not results from testing an assistant. The controls change this diagram only; they do not install a skill, connect a service or run an AI model.
Same task. Which part is missing?
Make a workshop brief from three lines of notes.
Compare what supplies the source and the method.
Prompt only: bring the notes and the method
You paste the three lines and request Decisions, Actions and Unknowns. No skill or file connection is needed for this version of the task.
Illustrative draft Decisions: workshop planned for Saturday, 10am. Actions: Mira to bring cards. Unknowns: room, calendar date and action deadline.
Skill: reuse the method, still supply the source
The workshop-brief skill carries the headings and the instruction to preserve unknowns. You still paste the notes. The folder does not give the assistant access to files elsewhere.
What changed? The reusable instruction supplies the format. It could produce the same kind of brief as the explicit prompt; a better answer is not implied.
Tool: retrieve the source, still specify the task
A configured file-reading capability can retrieve workshop-notes.txt. You request the brief and its format. The connection provides access; it does not establish the interpretation or a preferred writing method.
What to inspect The intended file and retrieved text, then the brief. This drawing does not configure or enforce file permissions.
Both: retrieve the notes and apply the saved method
The connection provides the source and the skill supplies the brief format. The two parts can complement each other. Missing room details still belong under Unknowns.
What did not change? More components have not confirmed the room. The example ends with a draft to review, rather than a message sent or a booking made.
A simplified teaching example with written sample outputs. No files are read, no tools are connected, and no AI runs in this illustration.
In this guide
First try a request with a visible finish
You could paste the three source lines and use the request below. Keep the notes beside the answer so you can compare each item. Naming a format in a prompt can accomplish the same formatting aim as a small skill; the reason to package it is reuse, rather than a new kind of intelligence.
An illustrative brief could record Saturday at 10am as the workshop plan, Mira bringing cards as the action, and the room and calendar date as unknowns. The notes do not establish a deadline for Mira or a particular Saturday. If an answer supplies those details, the source does not support them.
Turn the supplied workshop notes into a brief.
Use three headings: Decisions, Actions, Unknowns.
Keep names and times as supplied.
Put missing details under Unknowns rather than guessing.
Return the draft here for me to review.In this guide
Turn a repeated method into a small skill
If this becomes a weekly task, you could examine whether your app supports Agent Skills and where its current documentation says to add them. A folder location used by one client is not a universal installation path. Some products offer an interface; others discover files in documented locations.
The following starter is an instruction-only example. Save the content as SKILL.md inside a folder named workshop-brief if your chosen client supports that format and location. The name matches the folder; the description explains when the skill belongs. You could also keep the body as a saved prompt without installing it.
Before adding a third-party skill, consider reading its full contents, including referenced scripts and dependencies. A short description does not show everything it could ask the host to run. For this starter, there are no scripts, credentials or service connections.
After adding it through the client’s documented route, check that the app can find it and use the supported way to invoke it. Repeat the same notes and compare the format and missing-detail handling. A skill being listed and its instructions actually being used are separate observations.
---
name: workshop-brief
description: Prepare Decisions, Actions and Unknowns from supplied workshop notes. Use when the user asks for a workshop brief.
---
Use the source notes supplied for this task.
Organize the draft under Decisions, Actions and Unknowns.
Preserve supplied names and times.
List missing dates, rooms and deadlines as unknown.
Identify the source filename when it is available.
Return the brief for review.In this guide
Add a connection only when the task needs information or an action
Suppose the notes are saved in a file and you want the assistant to retrieve them. That is an access question, separate from remembering the brief format. The client needs a supported file-access capability; a skill mentioning a filename does not make that file available.
A common MCP setup has three pieces to identify: the client app, the server it connects to, and the capabilities the server exposes. Local servers run as programs in your environment; a remote connection reaches a service elsewhere. Installation requirements, authentication and available controls depend on the particular pairing.
The official local-server walkthrough uses a filesystem example with reading and writing capabilities. It is not a read-only guarantee. If you follow a concrete walkthrough, consider its current prerequisites and allowed locations, together with the client’s actual tool and approval settings. A request to read one practice file is not an enforced restriction on everything the setup can reach.
For a first connection, identify the exact source you need and whether the task needs reading, writing or another action.
You could start with a separate practice folder and replaceable notes, then inspect which tools and locations the client actually exposes.
Use the publisher’s current instructions for that client/server pair. A local process may require a runtime; a remote service may require a separate account or authorization.
Before running the task, locate the client’s connection status, available capabilities and removal controls. If access is unclear, a pasted sample still lets you practice the brief.
Ask for the source filename and the draft. Where activity is visible, compare the recorded retrieval with the file you intended.
In this guide
Inspect the new work, including what the addition did not change
The comparison is about what became available, not whether an answer sounds better. With the skill, you are inspecting the repeated method. With a connection, you are also inspecting the source retrieved and any actions attempted. With both, there are two parts to account for.
Perhaps change one line of the practice notes and repeat the task. If Saturday changes to Sunday, does the draft reflect the supplied revision? If the room remains unconfirmed, does it stay unresolved? This is a small freshness and fidelity check you can understand without inventing a score.
The visible draft and activity record answer different questions. A record can show that a read occurred without establishing that the model interpreted the file correctly. A well-formatted brief can still omit a condition or add an unsupported detail.
Source: which file, passage or pasted notes were available?
Method: was the requested structure and treatment of unknowns visible?
Activity: which operations are shown, and did anything exceed the planned task?
Correction: what did you change or discard before using the draft?
Fit: did the saved method or connection remove enough repeated work to keep it?
Takeaway
Keep the smallest addition that earns a role
For a one-off brief, a clear prompt and pasted notes may be sufficient. For a repeated format, a small skill may be worth exploring. For information that needs retrieving, a suitable connection may matter. Combining them makes sense when the task needs both; it does not supply facts absent from the source.
You could keep the note below alongside the practice source and accepted draft. When the app, skill or connection changes, it gives you a concrete task to revisit rather than starting again from its demo. There is no need to add another component just because it is available.
Task:
App and addition used:
Source and version tried:
Observed access or activity:
What I accepted or corrected:
What remains unknown:
Keep, remove or revisit — and why:
AI Guide note
How to use this guide
AI Guides are general editorial guidance, not professional advice or guarantees about accuracy, safety, suitability, performance, or outcomes. Tools, terms, prices, features, and laws can change. Check important details against original sources, product terms, reliable references, and qualified help where needed.
Source trail
Formats and connections behind the example
The skill format and MCP concepts below follow these documentation sources. The workshop, prompts and illustrated outputs are teaching examples.
What to explore next
Understand the system and its reach
Follow the model-and-agent explanation for the wider setup, or examine the access and review points when a connected tool can act on files or applications.