Getting Started Copy

Agents, Skills & MCP

Connect Claude Desktop, Cursor, Antigravity, and custom agent harnesses using the open Model Context Protocol (MCP) and custom skills.

BetterContext provides an enterprise-ready MCP server that exposes your organization's memory graph, tool connectors, and experience skills to any compliant AI agent harness.

Universal MCP Server Configuration

Add BetterContext to your agent harness configuration file:

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "bettercontext": {
      "command": "npx",
      "args": ["-y", "@bettercontext/mcp-server"],
      "env": {
        "BETTERCONTEXT_API_KEY": "bc_live_your_api_key_here"
      }
    }
  }
}

Cursor IDE (Features > MCP Servers):

Name: bettercontext
Type: command
Command: npx -y @bettercontext/mcp-server

Exposed MCP Tools

Once connected, agents gain access to these native capabilities:

  • bettercontext_query: Semantically and causally search the memory graph with optional role-based ACL filters.
  • bettercontext_upsert: Commit new architectural insights, incident findings, or code guidelines to persistent memory.
  • bettercontext_evaluate: Test candidate actions against Focus Mode causal inference models before execution.
  • bettercontext_search_skills: Discover and load organizational skills authored by other teammates.

Authoring Reusable Skills

Skills encapsulate repeatable procedural knowledge—such as PR review checklists, database migration runbooks, or deployment scripts. Create a SKILL.md in your project repository:

---
name: deploy-staging
description: Run automated staging deployments with pre-flight smoke tests
tags: [devops, k8s, staging]
---

# Deploy to Staging Runbook

1. Check that CI pipeline on `main` passed.
2. Ingest the release git tag into BetterContext: `bettercontext ingest --tag $TAG`.
3. Verify that environment secrets match current schema requirements.
4. Trigger staging rollout and monitor error rates for 5 minutes.

Sandboxing & Safety Guardrails

All tool executions requested by agents are evaluated against workspace security policies:

  • Read-Only Mode: Enforce read-only memory traversal for junior or external agents.
  • Human Confirmation Gates: Write operations, file modifications, or deployment triggers can be flagged for teammate approval.
  • Data Leakage Prevention: Proprietary code and environment secrets are masked before leaving your local machine.

Next Steps

Was this page helpful?

Powered By BetterContext