AI Agents & MCP

Give every AI tool your organization's brain.

75% of agentic AI tasks fail in production. The #1 cause? The agent had no context about your organization.

The problem

You paste the same context into every AI conversation. Your Claude project has a 200k token system prompt that's still not enough. Different team members give AI conflicting instructions. Every conversation starts from zero.

How it works

Here's how you build your knowledge graph step by step using the ai CLI.

Add your organizational context

Start with the knowledge that every AI interaction needs: who you are, what you do, how you work.

ai add "We are a B2B SaaS company serving mid-market \
  fintech companies. Our product is a payment \
  orchestration platform. We prioritize reliability \
  over speed-to-market. Our primary customers are \
  engineering and finance teams." \
  --title "Company Context" -t note

Set up workspaces

Create workspaces so AI loads only the context relevant to the current task.

ai workspace create "Engineering"
ai workspace create "Marketing"
ai workspace create "Client: Acme Corp"

Each workspace scopes AI context to specific domains and node types.

Build up domain knowledge

Add the specific knowledge that makes AI output useful — processes, decisions, policies, architecture.

ai add "Our API rate limits: 1000 req/min for standard, \
  5000 for enterprise. Rate limit headers: \
  X-RateLimit-Remaining, X-RateLimit-Reset. \
  Retry strategy: exponential backoff, max 3 retries." \
  --title "API Rate Limiting Policy" -t policy -d technology

Connect the MCP server

The MCP server gives Claude, Cursor, ChatGPT, and any MCP-compatible client direct access to your knowledge graph.

# In your Claude Desktop / Cursor MCP config:
{
  "mcpServers": {
    "apart": {
      "command": "npx",
      "args": ["@apart-tech/mcp-server"]
    }
  }
}

Any MCP-compatible AI tool can now search, read, and traverse your knowledge graph.

AI agents query your graph automatically

When you ask Claude or Cursor a question, they use the MCP tools to pull relevant context from your graph.

# In Claude or Cursor, just ask naturally:
"What are our API rate limiting rules?"
"How do we deploy to production?"
"What's our brand voice for enterprise clients?"

The AI searches your graph, traverses connections, and assembles a context package — no copy-pasting required.

The graph grows as agents work

AI agents can also add knowledge back to the graph. Decisions made during coding sessions, processes discovered during debugging — they all feed back in.

# Through MCP, an AI agent can:
ai add "Discovered that the payment webhook \
  needs idempotency keys to prevent duplicate \
  processing during network retries." \
  --title "Webhook Idempotency Requirement" -t decision

AI writes. Humans curate. The graph grows.

Every AI tool now has organizational memory

Your knowledge graph is the shared brain across all AI tools. Claude, Cursor, ChatGPT — they all access the same source of truth through MCP.

No more pasting context into every conversation

AI agents make decisions grounded in your actual processes

Workspaces ensure AI loads only relevant context

Knowledge compounds — every interaction makes the next one better

Start building your knowledge graph

Free during beta. No credit card required.