> [!WARNING]
> **Experimental:** MCP features for Crashlytics are Experimental, which means they aren't subject to any SLA or deprecation policy and could change in backwards-incompatible ways.
With Firebase Crashlytics MCP tools and prompts, you can interact with your
Crashlytics data using your MCP-capable AI-powered development tools, like
Antigravity, Claude Code, or Cursor. These Crashlytics MCP tools and prompts
provide critical context to your AI tool to help you manage, prioritize, debug,
and fix issues.
**After you [set up the Firebase MCP server](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#set-up-firebase-mcp-server),
you can use the MCP features for Crashlytics:**
[Guided workflow to prioritize and fix issues](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#guided-workflow-prioritize-and-fix)
[Free-form conversational debugging](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#free-form-conversational-debugging)
> [!IMPORTANT]
> **Important:** The currently available Crashlytics MCP tools and prompts are for use with apps that already use Crashlytics.
**What is MCP?**
[Model Context Protocol (MCP)](https://modelcontextprotocol.io/about)
is a standardized way for AI tools to access external tools and data sources.
*** ** * ** ***
## Before you begin: Set up the Firebase MCP server
This section describes the basic setup of the
[Firebase MCP server](https://firebase.google.com/docs/ai-assistance/mcp-server.md.txt) so that you can use
Crashlytics-specific MCP commands, prompts, and tools (which are all
described later on this page).
#### Prerequisites
Make sure your environment meets these requirements:
- A working installation of [Node.js](https://www.nodejs.org/) and
npm. Installing Node.js automatically installs the npm command tools.
- Your AI-powered development tool supports MCP integrations.
If you're working with a Unity project, review the
[FAQ about loading Crashlytics MCP tools and prompts](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#faq-crashlytics-mcp-tools-and-prompts-not-loading).
#### Configure your AI tool to use the Firebase MCP server
### Antigravity
To configure Antigravity to use the Firebase MCP server:
1. In Antigravity, click the menu in the Agent pane \> **MCP Servers**.
2. Select **Firebase** \> **Install**.
This automatically updates your `mcp_config.json` file, which you can view by
clicking **Manage MCP Servers** at the top of the MCP Store pane \>
**View raw config**, with the following content:
```json
{
"mcpServers": {
"firebase-mcp-server": {
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
}
```
### Claude
#### Claude Code
- **Option 1**: Install via plugin (Recommended)
The easiest way to set up the Firebase MCP server in Claude Code is to
install the official Firebase plugin:
1. Add the Firebase marketplace for Claude plugins:
```bash
claude plugin marketplace add firebase/firebase-tools
```
2. Install the Claude plugin for Firebase:
```bash
claude plugin install firebase@firebase
```
3. Verify the installation:
```bash
claude plugin marketplace list
```
- **Option 2**: Configure MCP server manually
Alternatively, you can manually configure the Firebase MCP server:
1. Run the following command under your app folder:
```bash
claude mcp add firebase npx -- -y firebase-tools@latest mcp
```
2. Verify the installation:
```bash
claude mcp list
```
It should show:
firebase: npx -y firebase-tools@latest mcp - ✓ Connected
#### Claude Desktop
To configure Claude Desktop to use the Firebase MCP server, edit the
`claude_desktop_config.json` file. You can open or create this file from the
**Claude \> Settings** menu. Select the **Developer** tab, then click
**Edit Config**.
```json
{
"mcpServers": {
"firebase": {
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
}
```
### Cursor
**Option 1: Marketplace plugin *(recommended)***
Install the
[Firebase plugin from the Cursor Marketplace](https://cursor.com/marketplace/firebase).
This automatically configures the MCP server and provides access to
[Firebase agent skills](https://firebase.google.com/docs/ai-assistance/agent-skills.md.txt).
**Option 2: One-click MCP setup**
If you only want to add the MCP server to your global configuration, click the
following button:
[](https://cursor.com/en/install-mcp?name=firebase&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsImZpcmViYXNlLXRvb2xzQGxhdGVzdCIsIm1jcCJdfQ%3D%3D)
**Option 3: Manual configuration**
If you prefer to configure the server for a specific project or want to edit
your settings manually, update your `mcp.json` file:
- For a specific project, edit `.cursor/mcp.json`
- For all projects (global), edit `~/.cursor/mcp.json`
```json
"mcpServers": {
"firebase": {
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
```
### Cline
To configure Cline to use the Firebase MCP server, edit the
`cline_mcp_settings.json` file. You can open or create this file by clicking
the MCP Servers icon at the top of the Cline pane, then clicking the
**Configure MCP Servers** button.
```json
{
"mcpServers": {
"firebase": {
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"],
"disabled": false
}
}
}
```
### VS Code Copilot
To configure a single project, edit the `.vscode/mcp.json` file in your
workspace:
```json
"servers": {
"firebase": {
"type": "stdio",
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
```
To make the server available in every project you open, edit your user
settings, for example:
```json
"mcp": {
"servers": {
"firebase": {
"type": "stdio",
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
}
```
### Windsurf
To configure Windsurf Editor, edit the file
`~/.codeium/windsurf/mcp_config.json`:
```json
"mcpServers": {
"firebase": {
"command": "npx",
"args": ["-y", "firebase-tools@latest", "mcp"]
}
}
```
In addition to the basic configuration described above for each AI tool, you can
[specify optional parameters](https://firebase.google.com/docs/ai-assistance/mcp-server.md.txt#options).
> [!NOTE]
> **Note:** When the Firebase MCP server makes MCP tool calls, it uses the same user credentials that authorize the Firebase CLI in the environment where it's running. Depending on the environment, this could be a logged-in user or [Application Default Credentials](https://cloud.google.com/docs/authentication/application-default-credentials).
*** ** * ** ***
## *(Recommended)* Guided workflow to prioritize and fix issues with `crashlytics:connect`
Crashlytics provides a guided workflow that's conversational and flexible
to help you prioritize and fix Crashlytics issues in your app. For example,
your AI tool can fetch issues, explain issues, identify potential fixes, and
even make the code changes for you.
This guided workflow is available via the `crashlytics:connect` MCP command.
> [!NOTE]
> **Note:** Not all AI tools support MCP commands, but you can still debug issues using [free-form conversation](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#free-form-conversational-debugging).
#### Access and use the command
1. If you haven't already,
[set up the Firebase MCP server](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#set-up-firebase-mcp-server), and then
start your AI tool.
2. Run the `crashlytics:connect` MCP command.
Most AI tools provide a way to conveniently access this workflow. If your
agent does not support MCP commands, we recommend using [free-form
conversational debugging](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp#free-form-conversational-debugging).
3. Use your AI tool to help prioritize and fix Crashlytics issues,
for example:
- View a list of prioritized issues.
- Debug a specific issue by providing its ID or URL.
- Request more information about a crash.
- Ask the agent its reasoning for a suggested root cause.
*** ** * ** ***
## Free-form conversational debugging
While we recommend [using `crashlytics:connect` for the best debugging
experience](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp#guided-workflow-prioritize-and-fix), you can also debug issues
using a free-form conversation with an AI tool that
has access to the Crashlytics MCP tools. This is especially important for
AI tools that don't yet support MCP prompts (often referred to as slash commands
or custom commands).
After you've [set up the Firebase MCP server](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#set-up-firebase-mcp-server),
try some of the following examples.
### Fetch an issue and crash context
When your AI tool has access to Crashlytics MCP tools, it can fetch critical
Crashlytics issue data like user and event counts, stacktraces,
metadata, and app version
information.
Here are some example prompts:
- `A customer reported an issue during login when using our latest release. What
Crashlytics issues do I have that could be related to this login trouble?`
- To answer this question, your AI tool will likely read your code to understand where login happens and use various Crashlytics MCP tools to retrieve issue data. Your AI tool will then try to determine whether an issue exists in the latest version that relates to the login flow.
- `The previous on-call engineer was investigating issue abc123 but wasn't able
to resolve it. She said she left some notes -- let's pick up where she left
off.`
- To answer this question, your AI tool will use various Crashlytics MCP tools to retrieve issue context and any notes posted to the issue. It might also fetch example crashes to resume the investigation into the issue.
### Document a debugging investigation
When debugging an issue, it's often helpful to maintain records for yourself or
your team. Crashlytics offers this capability in the Firebase console,
and your AI tool equipped with Crashlytics MCP tools can also help ---
for example: summarize an investigation, add a note with helpful metadata
(like a link to a Jira or GitHub issue), or close an issue after it's fixed.
Here are some example prompts:
- `Add a note to issue abc123 summarizing this investigation and the proposed
fix.`
- `We weren't able to get to the bottom of this issue today, summarize what we
learned and attach it to issue abc123 to pick back up later.`
- `Close issue abc123 and leave a note including the link to the PR that fixed
the issue.`
*** ** * ** ***
## Crashlytics MCP tool reference
The following tables list the Crashlytics MCP tools that are available via
the Firebase MCP server.
After you've [set up the Firebase MCP server](https://firebase.google.com/docs/crashlytics/ai-assistance-mcp.md.txt#set-up-firebase-mcp-server), your
AI tool can use these MCP tools to help you understand, debug, and manage
issues. These MCP tools are used in both the `crashlytics:connect` guided
workflow and in free-form conversations with your AI tool.
For the majority of use cases, these MCP tools are for LLM-use only and not for
*direct* use by a human developer. The LLM decides when to use these MCP tools
based on your interaction with your AI tool.
> [!NOTE]
> **Note:** You can also see this information using the command: `npx firebase-tools@latest mcp --generate-tool-list`
### Manage Crashlytics issues
The following table describes the tools available
to manage your Crashlytics issues.
| Tool Name | Feature Group | Description |
|---|---|---|
| crashlytics_create_note | crashlytics | Add a note to an issue from crashlytics. |
| crashlytics_delete_note | crashlytics | Delete a note from a Crashlytics issue. |
| crashlytics_update_issue | crashlytics | Use this to update the state of Crashlytics issue. |
### Fetch Crashlytics data
The following table describes the tools available
to get Crashlytics-related information about your apps.
| Tool Name | Feature Group | Description |
|---|---|---|
| crashlytics_get_issue | crashlytics | Gets data for a Crashlytics issue, which can be used as a starting point for debugging. |
| crashlytics_list_events | crashlytics | Use this to list the most recent events matching the given filters. Can be used to fetch sample crashes and exceptions for an issue, which will include stack traces and other data useful for debugging. |
| crashlytics_batch_get_events | crashlytics | Gets specific events by resource name. Can be used to fetch sample crashes and exceptions for an issue, which will include stack traces and other data useful for debugging. |
| crashlytics_list_notes | crashlytics | Use this to list all notes for an issue in Crashlytics. |
| crashlytics_get_report | crashlytics | Use this to request numerical reports from Crashlytics. The result aggregates the sum of events and impacted users, grouped by a dimension appropriate for that report. |
*** ** * ** ***
## Additional information
### How your data is used
Data governance is determined by the AI-powered development tool that you use,
and is subject to the terms defined by that AI tool.
### Pricing
Firebase doesn't charge you to use Crashlytics MCP tools and prompts or to
fetch Crashlytics data from our public API.
Any cost is determined by the AI-powered development tool that you use, and
could be determined by the volume of Crashlytics data that's used by the
AI tool. Note that Firebase doesn't offer an explicit way to control how much
data is loaded into context, but we do include sensible default guidance for the
model.
## Troubleshooting and FAQ
#### Crashlytics MCP tools and prompts aren't loading
The Firebase MCP server attempts to recognize when a codebase is using
Crashlytics by inspecting installed dependencies. This capability isn't yet
supported for Unity projects and doesn't cover some non-standard dependency
management systems for other platforms.
If Crashlytics MCP tools and prompts aren't loading for you, then as a
workaround, you can install the Firebase MCP server manually and use the
`--only crashlytics` argument to load Crashlytics MCP tools and prompts.