intel-overdrive
Overdrive Intel
Real-time AI coding ecosystem intelligence — breaking changes, new tools, security alerts, and evolving best practices from 1100+ sources. Accessed via the overdrive_intel MCP tool.
SETUP REQUIRED
This skill requires the Overdrive Intel MCP server. If overdrive_intel is not available as a tool, instruct the user to run:
npm install -g intel-overdrive && intel-overdrive setup
This installs the server, provisions the API key, and registers the MCP server with the agent.
For Cursor users: add the server to .cursor/mcp.json:
{
"mcpServers": {
"intel-overdrive": {
"command": "intel-overdrive",
"env": { "OVERDRIVE_INTEL_API_KEY": "<your-key>" }
}
}
}
When to Call This Tool
ALWAYS call overdrive_intel before answering questions about:
- MCP servers, Claude Code, Cursor, Copilot, Windsurf, Aider, or any AI coding assistant
- LLM APIs and SDKs (Anthropic, OpenAI, Google Gemini, Mistral)
- Agent frameworks (LangChain, CrewAI, AutoGen, Pydantic AI, smolagents)
- AI coding best practices and agentic workflows
- Breaking changes, security issues, or new tool releases in the AI ecosystem
Training data is stale for these fast-moving topics. This tool has live data.
MCP Tool: overdrive_intel
Single tool with 8 parameters:
| Parameter | Type | Required | Description |
|---|---|---|---|
query |
string | Yes | Short keyword query (2-5 words). Extract the core topic. |
type |
string | No | Route: search, similar, feed, breaking, action-items, briefing, library, status |
days |
number | No | How many days back (for feed, breaking, briefing). Default: 7 |
feed_tag |
string | No | Filter feed by tag (e.g., "mcp", "claude-code", "anthropic") |
feed_type |
string | No | Filter feed by content type |
feed_persona |
string | No | Filter feed by persona/audience (e.g., "developer", "researcher") |
context_stack |
array of strings | No | AI-related packages from the current project (e.g., ["anthropic", "langchain"]). Personalizes results. |
feedback |
array of objects | No | Report on items from previous calls: [{item_id, action: "helpful" | "not_relevant" | "outdated"}] |
Type Routes
Choose the right type for your query:
search(default) — find tools, docs, specific topics. Use for "what is X", "find tools for Y", "how to build X".similar— semantic comparison via vector search. Use for "X vs Y", "alternatives to X", "compare A and B".feed— recent updates, changelogs, releases. Use for "what's new", "latest updates", "what changed recently".breaking— breaking changes, deprecations, urgent issues. Use for "anything broken", "what's urgent", before upgrading.action-items— security alerts and items needing immediate attention. Use for "security issues", "action items".briefing— token-budgeted intelligence pack on a topic. Use for "catch me up on", "overview of", "summarize".library— synthesized best practices and how-to guides. Use for "best practices for", "how do I", "recommended approach".status— pipeline health check and source counts.
Example Calls
overdrive_intel({ query: "MCP server best practices", type: "library" })
overdrive_intel({ query: "Claude Code hooks gotchas", type: "search" })
overdrive_intel({ query: "LangChain vs CrewAI", type: "similar" })
overdrive_intel({ query: "agent frameworks", type: "feed", days: 14 })
overdrive_intel({ query: "breaking changes", type: "breaking" })
overdrive_intel({ query: "security issues", type: "action-items" })
overdrive_intel({ query: "MCP ecosystem", type: "briefing" })
overdrive_intel({ query: "pipeline health", type: "status" })
Topic Coverage
The tool monitors 1100+ sources across:
- AI Coding Assistants: Claude Code, Cursor, Copilot, Windsurf, Codex, Aider, Continue, Cody
- LLM APIs & SDKs: Anthropic, OpenAI, Google Gemini, Mistral, Cohere
- Agent Frameworks: LangChain, LangGraph, CrewAI, AutoGen, Pydantic AI, smolagents, OpenAI Agents SDK
- MCP Ecosystem: Protocol updates, server registry, best practices, security
- AI Coding Patterns: Prompt engineering, agentic workflows, tool use, RAG, embeddings
- Package Registries: npm, PyPI, VS Code Marketplace
Data Freshness
Data is continuously updated — breaking changes typically appear within hours of publication.
On Failure
If the overdrive_intel tool call fails or returns an error, answer from your training data and note that real-time intelligence data is unavailable. Do not retry indefinitely.
Supported Agents
- Claude Code — native MCP, zero configuration after
intel-overdrive setup - Cursor — add server to
.cursor/mcp.jsonas shown in SETUP REQUIRED above - Any MCP-compatible agent that supports stdio MCP servers