Integrate Spark with GitHub MCP

The GitHub MCP connector lets you query live GitHub data (open pull requests, CI/CD status, issue queues, and more) directly in Spark chat. Each user connects their own GitHub account, and the connector never writes to GitHub.

This article covers how to connect the GitHub MCP connector, what you can ask it, and how it behaves.

Note: Spark has a separate integration for codebase analysis, which indexes your repositories so Spark can answer questions about how your product behaves. There's also the GitHub Issues integration, which pushes features from Productboard into GitHub as issues. This article covers neither.

In this article:

• What it is
• Differences between the GitHub MCP connector and codebase analysis
• Connecting the GitHub MCP connector
• Using the GitHub MCP connector
• GitHub MCP behaviors
• FAQ
• See also

What it is

The GitHub MCP connector makes GitHub queryable in plain language, directly in Spark. Instead of switching to GitHub to piece together the state of a repository, you can ask questions like "What PRs are open in this repo?", "Did CI pass on the last commit?", or "Which issues are assigned to me?" and get a direct answer in chat.

The connector is read-only: it surfaces the current state of GitHub objects but never creates, edits, or deletes anything. 

Note: The GitHub MCP connector only works with GitHub.com. GitHub Enterprise (self-hosted) support is planned as part of a future Custom Connectors capability.

Each user connects their own GitHub account through Spark settings, so the connector reflects that user's individual permissions. It only accesses repositories the user already has permission to view in GitHub; no elevated access is granted.

MCP connector vs codebase analysis

These are two different types of integrations that answer different questions. Here's how to tell them apart:

If you have both tools set up, Spark will decide whether to use one or both depending on how your questions are worded. 

If you only have one or the other integration set up, Spark may struggle to answer certain questions. For example, if you only have the codebase integration set up, and you ask Spark about an open PR, it won't give you a good answer (because that question is better handled by the MCP connector.

See Analyze your codebase with Spark for more information about codebase analysis. 

Connecting the GitHub MCP connector

To connect your GitHub account to Spark:

1. From any Spark chat, click Integrations > Add integrations and connectors.
   
   
    
2. Find GitHub in the connector list and click Connect.
   
   
    
3. You'll be redirected to GitHub to authorize Spark. Review the permissions and grant access. When finished, you'll be redirected back to Productboard.

The connection is tied to your GitHub account and is active across all your Spark chats. Other workspace members connect their own accounts separately.

Using the GitHub MCP connector

Once connected, ask Spark about GitHub in natural language. For best results, name GitHub explicitly in your prompt so Spark knows where to look:

• "What PRs are open in the productboard/platform repo?"
• "Did CI pass on the last commit to main?"
• "Which issues are assigned to me this week?"
• "What's the current status of PR #1234?"
• "Show me open issues labeled bug in the api repo."

Spark queries GitHub on your behalf and returns a structured answer in chat. You stay in your workflow; no tab-switching required.

If GitHub is unreachable or returns an error, Spark surfaces a clear message and suggests checking GitHub directly.

GitHub MCP behaviors

• Read-only. The connector never writes to GitHub — no issues created, no PR comments posted, no branches modified.
• GitHub.com only. GitHub Enterprise Server (self-hosted) is not currently supported.
• Personal scope. The connector uses your individual GitHub credentials. It can only access repositories you already have permission to view.
• Questions about how your code actually works, like its architecture, business logic, or permissions, are better answered by codebase analysis instead. If your workspace doesn't have that set up, such questions will likely return unsatisfactory results. 

FAQ

See also

• Codebase analysis in Spark
• Connect external tools to Productboard Spark
• Productboard Spark
• Prompt library for common Spark use cases