Skills & MCP Servers: Extend Your AI Agents – Spedy Blog

Your Agents, Extended

Out of the box, Spedy AI agents can read code, write files, search the web, and post ticket comments. That covers a lot. But some tasks need more: a consistent output format, access to a specific API, or the ability to fetch content from an external service.

That's where Skills and MCP Servers come in.

Skills: Reusable Instructions for Agents

A Skill is a reusable instruction block — a short prompt that tells an agent exactly how to handle a specific type of task. You write it once, assign it to a pipeline stage, and every time that stage runs, the agent gets those instructions automatically.

What a Skill looks like

# Skill: Ticket Summary

## Purpose
Create a concise, structured summary of a ticket.

## Output Format
**Goal:** [One sentence — what the ticket wants to achieve]
**Approach:** [One sentence — how it should be implemented]
**Done when:** [One sentence — acceptance criteria]

## Rules
- Maximum 3 sentences total
- No bullet lists, no code blocks
- Stay neutral — don't add opinions or suggestions

When this skill is assigned to the Board Helper's "Respond" stage, every @Board Helper mention that asks for a summary gets exactly this format — no variation, no guessing.

Good uses for Skills

Ticket summaries — standardize how agents explain tickets to your team
Code review checklists — ensure every review checks security, performance, and test coverage
PR description templates — define the structure agents use when opening pull requests
Commit message conventions — enforce your team's commit style automatically
API documentation — give agents a concise reference for internal services they interact with frequently

How to set up a Skill

Go to Settings → Agent Skills
Click Add Skill — give it a name and write the instruction content
Open a Runner Team, select a pipeline stage
Under Skills, add the skill to the stage

The skill is now injected into every run of that stage. You can assign multiple skills to one stage — agents receive all of them.

MCP Servers: External Tools for Agents

MCP (Model Context Protocol) is an open standard that lets agents connect to external tools and data sources. Think of it as plugins for your AI agents — each MCP server exposes a set of tools the agent can call during a job.

Spedy supports two transport types:

HTTP/SSE — connects to a remote MCP server via URL
STDIO — runs a local process (e.g. an npm package) inside the runner container

Built-in MCP tools

Every Spedy runner includes built-in MCP tools out of the box:

Tool	What it does
`ticket_context`	Reads full ticket details, description, and metadata
`ticket_comments`	Fetches all comments on a ticket
`tickets_add_comment`	Posts a comment on a ticket
`tickets_update_description`	Rewrites a ticket's description
`knowledge_recall`	Retrieves stored knowledge entries
`knowledge_search`	Searches the knowledge base semantically
`knowledge_store`	Saves a new knowledge entry
`clone_repo`	Clones a repository for code context

These are always available — no configuration needed.

Adding external MCP servers

Go to Settings → MCP Servers and click Add MCP Server. You can add:

Fetch MCP — lets agents fetch URLs and return their content:

Command: npx
Arguments: -y fetch-mcp
Exposes: fetch_url, fetch_youtube_transcript

GitHub MCP — gives agents read/write access to GitHub:

Command: npx
Arguments: -y @modelcontextprotocol/server-github
Set env var: GITHUB_TOKEN

Postgres MCP — lets agents query your database:

Command: npx
Arguments: -y @modelcontextprotocol/server-postgres
Set env var: DATABASE_URL

Slack MCP — agents can post to Slack channels:

Command: npx
Arguments: -y @modelcontextprotocol/server-slack
Set env var: SLACK_TOKEN

Assigning MCP servers per stage

Once added, you assign MCP servers to individual pipeline stages. A "Respond" stage for the Board Helper might only need fetch-mcp. A "Code Review" stage doesn't need it at all. Keep each stage's tool surface minimal.

Open a Runner Team → select a pipeline stage
Scroll to MCP Servers
Toggle on the servers this stage should have access to

Example: Board Helper with Fetch

Assign fetch-mcp to the Board Helper "Respond" stage and mention the agent in a ticket:

@Board Helper fetch https://docs.stripe.com/api/payment_intents and summarize the key fields

The agent calls mcp__Fetch__fetch_url, gets the content, summarizes it, and posts the result as a ticket comment — all without leaving Spedy.

Combining Both

Skills and MCP Servers work best together. A skill defines how the agent should respond; an MCP server gives it the data to respond with.

Example setup: Research Assistant stage

Skill: forces the output format (e.g. bullet summary + source URLs)
MCP Server: fetch-mcp to retrieve external pages
Allowed Tools: Read, Glob, mcp__Fetch__fetch_url, mcp__spedy__tickets_add_comment

Result: a focused, reliable research agent that always delivers consistent output.

What's Next

Skills and MCP Servers are the foundation for agent specialization. Next up: a skill marketplace with community-contributed templates, and support for authenticated HTTP MCP servers with per-org credentials.