Meta Developer Tools MCP: Connect AI Agents to Your App Dashboard

Meta Developer Tools MCP: Connect AI Agents to Your App Dashboard

Meta’s Developer Tools MCP is a beta remote server that connects compatible AI coding clients to the Meta developer platform. It lets an agent discover documentation, inspect app configuration, check App Review and compliance status, review API health, and inspect or manage webhook subscriptions without switching to the developer dashboard.

The practical boundary matters: almost all capabilities are read-only. Manage scope adds write access only for webhook subscriptions. This guide explains the architecture, setup, permissions, security model, troubleshooting workflow, and when the server is useful.

The short answer

Meta Developer Tools MCP is a Streamable HTTP MCP server at https://mcp.facebook.com/devtools. It authenticates with Meta OAuth and exposes ten namespaced tools through compatible clients such as Claude, Cursor, Codex, and ChatGPT.

Use it when an engineer needs platform context while working on an integration: the agent can look up the right Meta documentation, inspect an app’s configuration, find compliance actions, check API usage and deprecations, or diagnose webhook configuration. Do not treat it as a general-purpose deployment agent. The beta’s write surface is intentionally narrow and limited to webhook subscription operations.

What Meta shipped

Meta announced Developer Tools MCP on June 30, 2026, describing it as a way to explore and debug Meta app integrations from an AI agent. The companion documentation, updated June 12, identifies the server as beta, says rollout is gradual, and warns that the interface and tool set may change.

The server uses:

• Transport: Streamable HTTP
• Endpoint: https://mcp.facebook.com/devtools
• Authentication: Meta developer account sign-in through OAuth
• Status: Beta
• Scopes: Read and Manage, granted per connected app

The server is not the same as Meta’s advertising connectors. It targets developers building apps and integrations on the Meta platform, not campaign operators managing ads.

Tools and permissions

The ten tools fall into five useful groups:

The documentation is explicit about the most important security fact: Manage scope adds only the ability to create, update, and delete webhook subscriptions. It does not let the agent change arbitrary app settings, submit App Review, or alter compliance status.

That is a sensible default for an agent connector. Read access can still expose sensitive configuration, usage patterns, compliance findings, and webhook details, so “read-only” should not be confused with “risk-free.”

Architecture

The agent client is the control surface, Meta’s remote MCP server is the protocol boundary, and the developer’s selected apps are the authorization boundary.

This design also fits the broader MCP deployment direction covered in the MCP 2026-07-28 stateless-server migration guide: a remote HTTP server is easier to place behind normal infrastructure than a local process, but the application still needs explicit authorization, state, logging, and retry rules.

Setup in supported clients

Claude Code

Add the remote server:

1claude mcp add --transport http meta_developer_tools https://mcp.facebook.com/devtools

In a session, run /mcp, select meta_developer_tools, and complete the OAuth flow in the browser.

Claude Desktop

Open Settings → Connectors → Add custom connector. Set the name to Meta Developer Tools and use:

1https://mcp.facebook.com/devtools

Codex App and ChatGPT

For the Codex App, open Settings → MCP Servers → Add servers, choose Streamable HTTP, enter the endpoint, select OAuth, and authenticate. The documentation says to restart the Codex App and verify the server entry afterward.

For ChatGPT, first enable Developer Mode under Settings → Connectors → Advanced settings. Then create a connector with the same endpoint and OAuth authentication. The connector appears in the composer’s Developer mode menu.

Cursor

Cursor reads remote MCP servers from mcp.json. A global configuration applies to all projects; a project-level .cursor/mcp.json limits the connection to one repository.

1{
2  "mcpServers": {
3    "Meta Developer Tools": {
4      "url": "https://mcp.facebook.com/devtools",
5      "type": "http"
6    }
7  }
8}

Cursor should show a Needs login action after the file is saved. Authenticate before asking the agent to inspect an app.

OAuth and app authorization

The connection does not require copying an App ID or App Secret into an MCP configuration file. The client starts an OAuth flow, the developer signs in with Meta, and the consent screen lets the developer choose which apps the server may access.

The documentation says the sign-in flow must be completed again when the client restarts. That has an operational consequence: teams should document the expected authentication step instead of assuming a long-lived unattended connection.

Scopes are configured per app in Meta’s Business Integrations settings:

• Read: app configuration and settings, App Review status, compliance status, API usage and health, and webhook topics and subscriptions.
• Manage: everything in Read, plus webhook subscription creation, updates, and deletion.

Start with Read. Grant Manage only to a tightly controlled development or staging app when webhook changes are genuinely needed.

A verification-first workflow

Do not begin by asking an agent to change anything. Verify the connection and the returned context in stages.

1. Confirm tool discovery

Ask the client to list the tools available from Meta Developer Tools. You should see the devtools_ namespace and ten tools. If the list is empty, stop and fix authentication or endpoint configuration before proceeding.

2. Run a low-risk read

Use devtools_app_list to identify accessible apps, or devtools_api_changelog to confirm that documentation-level tools work without app-scoped access.

3. Inspect one app narrowly

Ask for one specific read operation, such as the basic settings or current App Review status of a known staging app. Require the agent to identify the app ID and distinguish returned facts from recommendations.

4. Cross-check important conclusions

For a compliance deadline, permission status, deprecation, or rate-limit concern, compare the agent’s answer with the relevant Meta dashboard or source documentation. The agent is a faster interface, not an independent authority.

5. Test webhooks only after reads are trusted

List topics and subscriptions first. A webhook test requires an active subscription and a valid topic/field combination. For subscription changes, verify that the callback is a live HTTPS endpoint and that the operation is being performed against the intended app.

This staged approach mirrors the safer patterns in the Agent Plugins 1.0 guide: a connector’s package or protocol may be portable, but approvals, permissions, provenance, and sandboxing remain client and operator responsibilities.

Security and privacy considerations

Treat app configuration as sensitive

App settings, platform restrictions, review history, compliance findings, usage data, and webhook subscriptions can reveal operational and business information. Keep prompts and transcripts inside approved workspaces, and check whether the selected client stores tool output remotely.

Separate development and production access

Use a staging app for experiments. Do not grant Manage scope to every developer or every client. If production access is required, use the smallest permitted scope, a named owner, and an audit trail for tool calls.

Put a human between diagnosis and change

An agent can identify a missing subscription or propose a correction, but a human should confirm the app, topic, fields, callback URL, and expected impact before a write operation. The narrow write surface reduces risk; it does not remove confirmation requirements.

Avoid secret leakage

Do not paste App Secrets, access tokens, webhook signing secrets, or private customer data into prompts. OAuth avoids putting an App Secret in mcp.json, but the agent may still see sensitive data returned by an authorized tool.

For a more general view of agent execution boundaries, compare this connector with the harness engineering guide for AI coding agents. A reliable harness should make permissions, evidence, stop conditions, and review explicit rather than relying on the model to remember them.

Common errors and debugging

“It looks like this app isn’t available”

Meta says this means the account does not yet have approved access to the beta. The rollout is gradual, so retrying the same prompt will not solve an access decision.

“Facebook login is currently unavailable for this app”

The client may not be supported yet, or the connector configuration may not be using the documented OAuth path. Confirm that the endpoint is exactly https://mcp.facebook.com/devtools and update the client before debugging the app itself.

The client connects but exposes no tools

Restart the IDE or agent client, authenticate again, and ask it to list tools. Then run devtools_app_list or devtools_api_changelog. A missing tool list is a connection problem, not evidence that the app has no configuration.

Webhook management fails

Check all of the following:

• The selected app has Manage scope.
• The callback URL is live HTTPS and passes Meta verification.
• The topic and fields are valid for the app.
• The intended app ID is being used.
• The agent is not attempting an unsupported app-setting change.

The answer seems stale

Meta labels the service beta and says the interface and tool set may change. Re-run the read operation, inspect the dashboard, and consult the current Meta documentation or changelog before making a release decision.

When should developers use it?

Developer Tools MCP is a good fit for teams that repeatedly switch between an editor, Meta documentation, App Review screens, compliance notices, and webhook settings. Its highest-value use cases are context retrieval and diagnosis: finding endpoints, understanding errors, checking readiness, and summarizing app health.

It is a weaker fit for unattended production automation, bulk administration, or workflows that require broad write access. For those cases, use narrowly scoped service integrations and conventional deployment controls, then let an agent assist with evidence and review.

FAQ

Is Meta Developer Tools MCP an API replacement?

No. It is an agent-facing connector to Meta’s developer platform. Your application still uses Meta’s documented APIs, SDKs, webhooks, and authentication flows.

Can it change my entire Meta app configuration?

No. According to Meta’s documentation, Manage scope adds write access only for creating, updating, and deleting webhook subscriptions. Other exposed app operations are read-only.

Does it work with every MCP client?

No. Meta documents setup for Claude, Codex, ChatGPT, and Cursor, and says availability is gradual. Client support and behavior may change while the service is beta.

Do I need an App ID or App Secret in the client configuration?

No. The documented setup uses OAuth with a Meta developer account. You still need the appropriate access to the app and must choose which apps to authorize.

Is it safe to grant Manage scope?

Only when the app, client, operator, and callback endpoint are controlled. Start with Read, use staging, require confirmation, and log every write operation.

Conclusion

Meta Developer Tools MCP is a useful example of where remote MCP connectors are becoming practical: not as magical autonomous operators, but as authenticated context layers inside the developer workflow. It puts documentation search, app health, review status, compliance, API usage, and webhook diagnostics near the code that depends on them.

The safest adoption path is straightforward: connect with Read scope, verify tool discovery, inspect a staging app, cross-check important facts against Meta’s source pages, and enable Manage only for deliberate webhook work. The protocol makes the interface convenient; engineering discipline still determines whether the integration is trustworthy.

Sources

• Meta for Developers: Announcing Developer Tools MCP
• Meta Developer Documentation: Developer Tools MCP
• PPC Land: Meta launches Developer Tools MCP

Visual: original Mermaid architecture diagram by Essa Mamdani; no external image used.

Visual: Integration request flow

This original architecture diagram shows how the components described in this article fit together. It is a practical reference for deciding where authentication, validation, retries, and observability belong.

Visual reading: keep the client, policy boundary, external service, and result validation separate. This prevents an AI-generated tool call from becoming an unchecked side effect.