Connect Claude, Cursor, and any AI assistant to your data through MCP or CLI. Build custom skills and agents that automate workflows, answer questions, and extend your docs with code.
MCP-compatible • Works with Claude Desktop, Cursor, and more
Two powerful concepts that work together
Single-purpose tools
A skill is a reusable function that performs one specific task. Think of it as a building block.
Orchestrate multiple skills
An agent is an intelligent workflow that combines multiple skills to accomplish complex tasks autonomously.
Agents orchestrate skills to accomplish complex tasks
An agent knows what to do. An identity defines how it sounds. Attach an identity to a workflow and every agent inside it writes with a consistent voice, tone, and persona — automatically.
A specific individual's voice — useful for ghostwriting, personal brand content, or an agent that speaks as a named expert on your team.
Your brand voice — tone of voice guidelines, style rules, and communication principles encoded once and applied by every agent across every output.
A blend of person and brand — ideal for a founder who wants outputs that feel personal but stay on-brand, or a team persona that mixes individual warmth with company style.
Attach an identity to a workflow and every agent in that workflow inherits it. No per-agent configuration needed.
"The error you encountered is caused by an invalid API key. Please verify your credentials in the dashboard settings and retry the request."
Generic, robotic — could have come from anywhere.
"Hey! Looks like your API key might be incorrect. Head over to your dashboard settings to double-check — should be a quick fix. Let us know if you're still stuck!"
On-brand, warm, recognisably you.
Model Context Protocol (MCP) integration in 60 seconds
Model Context Protocol is a universal standard that lets AI assistants (Claude Desktop, Cursor, etc.) connect to external data sources—like your Contextium documentation.
Contextium runs a hosted MCP server. Just paste a short config snippet into your AI tool—no packages, no dependencies.
"Based on your API Security Guide (tagged #api #security):
1. Client requests access token using API key
2. Server validates key and issues JWT token
3. Client includes token in Authorization header
4. Token expires after 24 hours, refresh required
Source: API-Security-Guide.md (updated 2 days ago)"
Claude searched your docs, found the right document, and gave you the exact answer—all in one conversation.
Start with our library, then build your own
See how teams use agents to automate documentation tasks
Generates personalized onboarding documentation for new team members in seconds.
Answers customer questions from your docs, creates tickets for unresolved issues.
Scans all documentation weekly, finds outdated content, notifies owners to update.
Easy-to-use SDK with full documentation
import { Skill } from '@contextium/sdk';
export const mySkill = new Skill({
name: 'generate-summary',
description: 'Generates a summary',
async execute(docs) {
// Your logic here
const summary = analyzeDocs(docs);
return summary;
}
});Deploy to Contextium and it's instantly available to all your agents and AI assistants.
import { Agent } from '@contextium/sdk';
export const myAgent = new Agent({
name: 'onboarding-bot',
async run(input) {
// Chain multiple skills
const template = await search();
const doc = await create(template);
await tag(doc, ['onboarding']);
await notify(input.user);
}
});Agents can orchestrate multiple skills, make decisions, and run on schedules or triggers.
Start with pre-built skills, connect your AI, and build custom agents for your unique workflows.
Skills & Agents are one layer of a complete AI collaboration platform
Each workflow specifies which agents to load at session start. Your code review agent loads for the Engineering workflow, your content agent for Marketing — the right specialist is always in context, automatically.
Learn about WorkflowsThe three core project roles are agents. The Planner writes the spec. The Builder implements phase by phase. The Reviewer grades before anything is marked complete. Build them once, reuse them across every project in your workspace.
Learn about ProjectsWhen an agent loads a workflow, it reads the active task list — status, priority, assignee, and due date. It picks up in-progress work first, then moves to the highest-priority to-do. No briefing needed between sessions.
Learn about TasksOnce your agent is polished, publish it to the Contextium Marketplace for other teams to use. Browse hundreds of ready-made specialist agents — or install community-built ones into your workspace in a single click.
Learn about the Marketplace