Connect a custom MCP server to Unblocked to bring your own data source

Every team runs something Unblocked doesn’t have a built-in data source for — an internal service, a proprietary API, or a platform like PagerDuty, Grafana, Backstage, LaunchDarkly, or Splunk. Custom MCP is how you bring it in yourself: connect any Model Context Protocol server that is reachable over public HTTPS, and Unblocked uses its tools when they are relevant to a question.

Custom MCP is a pull model. Nothing is ingested or indexed — Unblocked calls the tools you selected at the moment the question is asked, so the answer reflects the current state of the system instead of your last upload. If the content behaves more like a document set you can hand over on a schedule, push it through the Unblocked API instead; the docs lay the trade-offs out side by side under Choose a Public API Collection.

Only read-only tools are ever available. Unblocked honors the readOnlyHint annotation your server provides, and where the hint is missing it evaluates the tool description and excludes anything that might modify data — then you choose which of the remaining tools Unblocked can use.

Once connected, any MCP-compatible coding agent — Claude Code, Cursor, Devin, GitHub Copilot, Codex — reaches that context through a single Unblocked MCP server connection, with no per-agent configuration for your server.

What you get

• Bring your own integration — Connect an internal tool, a proprietary API, or a platform Unblocked doesn’t support yet, without waiting for a built-in data source.
• Live context, nothing indexed — Unblocked calls your server while it answers, so results reflect the system as it is at that moment rather than a snapshot.
• Read-only by construction — Tools that can modify data never appear for selection: Unblocked honors annotations.readOnlyHint and otherwise excludes tools whose descriptions suggest a write.
• Tool-level control — Pick exactly which discovered tools Unblocked may use, and change the selection anytime from Settings → Data Sources.
• OAuth or HTTP header authentication — Authorize through your server’s OAuth flow, or have Unblocked encrypt the header values you supply and send them with every request.
• Cross-source synthesis — Tool results are reconciled with your code, pull requests, issues, incidents, docs, and conversations in one knowledge graph.

Need screenshots or troubleshooting?This is a concise connection summary. Follow the complete setup guide

Prerequisites

• An Unblocked account (free to start).
• The publicly reachable HTTPS endpoint of the MCP server, including its MCP path where one is required (for example https://mcp.example.com/mcp). Local, private-network, and plain-HTTP endpoints are not supported.
• An MCP server that supports tool discovery and gives its tools clear, accurate descriptions — hosted by a third party or operated by your team.
• For header-based authentication, the header names and values the server expects, such as an API key or a bearer token. Optionally, a public image URL to override the generated data source icon.
• An MCP-compatible coding agent (Claude Code, Cursor, Devin, etc.) to use the context in your editor.

Step 1: Connect your MCP server to Unblocked

1. In Unblocked, go to Settings → Data Sources → Connect another data source, select Custom, then select Custom MCP.
2. Enter the MCP server URL, adjust the generated name and icon URL if you want something more recognizable.
3. Choose how Unblocked authenticates: OAuth, which redirects you to the server to authorize access, or HTTP Headers, adding a name and value for each header the server requires.
4. Click Connect. Unblocked validates the endpoint, connects to the server, and discovers its tools — discovery can take up to a minute.
5. Select the read-only tools Unblocked may use, then click Save Settings.

Use a dedicated, least-privilege credential that grants access only to the data the selected tools need. For bearer tokens, use Authorization as the header name and Bearer YOUR_TOKEN as its value.

If Unblocked already has a built-in data source for the system — Datadog or Sentry, for example — connect that instead. Built-in sources are indexed, kept in sync, and cited in answers, so a Custom MCP passthrough to the same system gives you less.

Step 2: Install the Unblocked MCP server

On macOS or Linux, one command auto-detects and configures the supported agents on your machine:

On Windows, or when you prefer remote MCP, add the server manually using the configuration below, or follow the manual install guide .

claude mcp add -s user --transport http unblocked https://getunblocked.com/api/mcpsse

{
  "mcpServers": {
    "unblocked": {
      "url": "https://getunblocked.com/api/mcpsse"
    }
  }
}

https://getunblocked.com/api/mcpsse

Tool names come from the server you connect, not from Unblocked. When you click Connect, Unblocked discovers your server’s tools, excludes any that might modify data, and lists the read-only ones for you to select. Clear names and descriptions are how Unblocked decides when to call each tool. For example, a read-only log search tool on your server might appear as:

• search_logsSearch application logs — annotated with readOnlyHint, so Unblocked offers it for selection. Your server’s own tools replace this example when you connect.

To give Claude Code, Cursor, GitHub Copilot, Codex, or another MCP-compatible AI coding agent access to your MCP server, connect the server to Unblocked and add the single Unblocked MCP server to your agent.

Unblocked acts as the MCP context layer for your server, calling the tools you selected alongside your other connected sources and returning cited results through these tools:

• context_researchResearches a question across your connected sources and calls the selected tools on your custom MCP server when they are relevant, returning a synthesized answer with citations.
• context_get_urlsFetches the full content behind URLs your agent encounters across connected sources — Jira, Linear, Notion, Confluence, Slack, GitHub, and more.

Questions you (or your agent) can ask Unblocked once your MCP server is connected:

• What does our internal deployment tool report for the checkout service right now?
• Which on-call rotation owns this alert, and what code does it point at?
• What do the Grafana dashboards show for error rates since this morning?
• Is the feature flag for the new billing flow enabled in production?
• What does our internal inventory service say about this customer’s plan?