Meta this week rolled out the Developer Tools MCP server, a beta system that lets AI coding assistants read and, in one narrow case, modify a developer's Meta app configuration without the developer ever opening the Meta for Developers dashboard. The company announced the launch through a post on the Meta for Developers blog written by Zoë Lieberman, alongside a parallel documentation page describing setup steps for four supported clients: Claude Code, Claude Desktop, the OpenAI Codex App, ChatGPT, and Cursor.

The server sits at a single endpoint, https://mcp.facebook.com/devtools, and speaks over Streamable HTTP rather than the older stdio transport that many local MCP servers still use. Authentication runs through OAuth tied to a developer's existing Meta developer account, which means no App ID or App Secret needs to be copied into a client configuration file. According to Meta's developer documentation, the status of the server is listed plainly as Beta, with a note that "the interface and tool set may change." A separate banner on the same page warns that the rollout is gradual and "may not be available to everyone yet."

What the server actually does

Ten tools sit behind the devtools_ namespace, and the documentation groups them loosely by function rather than by client. Two of the ten, devtools_discovery and devtools_api_changelog, work without any app-scoped permission at all; a developer can search Meta's developer documentation or look up changelog products and RSS feeds before even connecting an app. The remaining eight require the developer to have granted either a Read or a Manage scope on a specific app, set through Business Integrations settings.

devtools_app_list is designed as a starting point. It returns every app a developer, admin, or tester can access through the server, along with the permission level granted on each. That app_id is then the input every other app-scoped tool needs. devtools_app inspects basic settings, advanced settings, security configuration, platform restrictions, and data protection officer details for a given app. devtools_app_review checks the status, history, required privileges, and submission requirements for App Review, the process by which Meta approves an app's use of specific permissions.

Compliance gets its own tool. devtools_compliance surfaces open required actions, violations, and recommendations for an app, which according to the announcement post is meant to solve a specific recurring problem: developers finding out an app lost access only after missing an email about a Data Access Renewal deadline. The blog post frames this directly, stating that the tool means "your agent proactively flags what needs attention and when action is required," rather than a developer discovering the lapse after the fact.

Two tools handle operational health. devtools_api_usage checks rate limits, call volume, and API deprecations an app depends on. devtools_webhook_list lists the webhook topics available to an app and the subscriptions already configured on it, including which fields are subscribed. The remaining pair, devtools_webhook_manage and devtools_webhook_test, are the only tools that touch write access. Creating, updating, or deleting a webhook subscription requires the Manage scope specifically; subscribing an app to a new topic also requires a live HTTPS callback URL that passes Meta's own verification step before any events will flow. devtools_webhook_test then sends a sample payload to a configured subscription so a developer can confirm the receiving endpoint behaves correctly, provided the field being tested already belongs to an active subscription.

Every other read across the ten tools - app settings, App Review status, compliance findings, rate limits, webhook topics - stops at Read scope. Manage scope, by the documentation's own description, adds exactly one additional capability: write access to create, update, and delete webhook subscriptions. Nothing else in the tool set can alter a developer's app configuration, submit an App Review request on a developer's behalf, or change compliance status.

Setup differs by client, but OAuth is constant

Configuration steps vary across the four supported integration paths, though the underlying authentication model does not. In Claude Code, a developer runs a single command, claude mcp add --transport http meta_developer_tools followed by the endpoint URL, then authenticates by running /mcp in a session and selecting the new server. Claude Desktop takes a slightly different route through its settings menu, where a developer manually adds a custom connector with a name and URL rather than a terminal command.

The OpenAI Codex App and ChatGPT both route through a Settings > Connectors path, though ChatGPT requires an additional step: enabling Developer Mode under advanced connector settings before the option to add a custom connector even appears. Once configured, according to the documentation, "Meta Developer Tools appears in the composer's Developer mode menu during conversations." Cursor reads its MCP configuration from a shared mcp.json file - either a global copy at ~/.cursor/mcp.json or a project-specific one - which the Cursor app, the Cursor CLI, and any other Cursor-based tool all reference identically.

Regardless of which client a developer picks, the sign-in flow itself follows four steps, according to the documentation: starting the connection, which some clients trigger automatically when the server is added and others require a manual Connect click; signing in through a browser window to the developer's Meta account; selecting, on a consent screen, which specific apps to grant the server access to; and confirming the connection back in the client. That sign-in has to be repeated each time the client restarts, since the documentation notes no persistent session survives a client restart on its own.

Two error states are documented explicitly. If a developer's account lacks approved access to the beta, the client returns the message "It looks like this app isn't available." If the specific MCP client in use isn't yet supported, the error reads "Facebook login is currently unavailable for this app." Meta's troubleshooting guidance for a working connection is a three-step check: restart the client, ask the agent to list its available tools, and run a low-risk read call such as devtools_app_list before attempting anything else.

The problem Meta says it is solving

The blog post describing the launch opens with a description of a familiar workflow friction point for anyone building on Meta's APIs. According to Meta, developers routinely lose time "searching docs to find the right endpoint, checking your developer dashboard for required actions, trying to figure out why an API call is failing," and each of those steps pulls a developer's attention away from the code editor they are actually working in.

The announcement organizes the server's value around five practical scenarios rather than around the ten tools individually. API discovery through conversation means a developer can describe an integration goal in plain language - the post's own example is "Help me setup an integration for Whatsapp business" - and receive back relevant Graph API endpoints, required permissions, and code samples in the language the developer is already using. Troubleshooting with platform intelligence means pasting a raw API error into the agent and getting back a resolution path informed by "live platform logic, changelog history, and known error patterns," rather than a developer manually cross-referencing an error code against changelog entries by hand.

App status and health visibility covers the devtools_app and devtools_api_usage functions together: a developer can check settings, platform details, API usage, and health status "without logging into the Dev Center," which according to the post lets a developer "quickly determine whether a problem is in your code or your platform setup." Required actions status at a glance is the framing Meta gives specifically to the compliance tool, and it is the one scenario in the announcement pinned to a concrete risk: a missed email leading to lost app access. Webhook debugging closes the list, covering the three webhook-specific tools together as a way to "get visibility into subscription status and event configuration all in one place."

Positioning within Meta's broader MCP strategy

The Developer Tools MCP server is a distinct system from the Ads AI Connectors that Meta launched on April 29, 2026, though the two share an underlying protocol. Meta opened its ad system to Claude and ChatGPT through that earlier release, which paired an ads-specific MCP server with a companion command-line tool, giving external AI agents write access to live campaign creation, ad set management, and product catalog operations inside a real advertising account. That system targeted advertisers and media buyers directly. The Developer Tools MCP server targets a separate audience: the engineers and technical teams who build the apps and integrations that sit on top of Meta's platform in the first place, rather than the people running ad campaigns through those apps.

The distinction matters because of what each system is permitted to touch. Meta's Ads CLI, published the same day as the ads connectors, allowed campaigns, ad sets, ads, and creatives to be created, listed, updated, and deleted directly from a command line or an AI agent. The Developer Tools MCP server, by contrast, keeps nearly its entire surface area read-only, with webhook subscription management as the sole write capability across all ten tools. That is a materially more conservative posture than the write-heavy ads connectors launched roughly two months earlier, and it places Meta's developer-facing MCP implementation closer to the read-only pilot that Microsoft ran for its own Advertising MCP server, which expanded to open pilot on June 17, 2026 with access limited to querying live campaign data rather than modifying it.

Model Context Protocol itself is an open standard that Anthropic introduced in November 2024 and later donated to the Linux Foundation. It defines a consistent way for AI applications to query external tools and data sources, functioning as what multiple platform announcements have separately described as a standardized connector layer between a model and whatever system it needs information from. Since its introduction, the protocol has moved from a developer-tooling curiosity into infrastructure adopted across the advertising industry: Amazon Ads ran a closed beta for its own MCP server starting November 13, 2025, supporting connections from Claude, ChatGPT, Amazon Q, and Amazon Bedrock, while Google explored a comparable server for its Ads API as far back as July 7, 2025, later shipping an open-source read-only version.

The compliance-visibility framing built into the Developer Tools MCP server also lands against a specific regulatory backdrop for Meta's developer platform. Meta consolidated what had previously been five separate compliance processes - data handling questions, Data Use Checkup, App Review, Data Protection Assessment, and ongoing reviews - into a single annual Data Access Renewal system, which began rolling out in phases starting October 2024. That consolidation was explicitly designed to reduce the number of separate deadlines a developer needed to track across a year. The new devtools_compliance tool extends that same logic one step further: instead of a developer needing to check a renewal status manually, the status becomes something an AI agent can surface unprompted inside the developer's own coding environment.

What remains unclear

Several details are left open in Meta's own documentation. The announcement does not specify a target date for when the beta might reach full availability, stating only that tool availability "is limited and subject to change." Nor does the documentation quantify how many developers currently have approved access, or what criteria determine which accounts are admitted to the beta ahead of others. The blog post links to a feedback form embedded within the tool itself, suggesting Meta is actively soliciting input before finalizing the tool set, consistent with the "interface and tool set may change" language attached to the server's beta status.

The scopes system, while described in detail, also leaves some operational questions unanswered. Read and Manage scopes are both granted per app rather than globally, meaning a developer with access to multiple apps could plausibly hold Manage scope on one and only Read on another, though the documentation does not describe how an agent behaves if it attempts a write action against an app where only Read has been granted. The documentation states scopes and access can be revoked at any time through Business Integrations settings, giving developers a way to walk back access without needing to remove the MCP server connection entirely.

Why this matters for the advertising technology community

For PPC Land's readership - largely people who manage advertising accounts, technology stacks, and API integrations across Meta's platform - the Developer Tools MCP server sits one layer below the tools most readers touch daily. It does not manage ad campaigns, adjust bids, or interact with the Marketing API's advertising objects the way the Ads AI Connectors and Ads CLI do. It manages the plumbing that sits underneath any custom integration a marketing technology team has built on top of Meta's platform: webhook subscriptions that feed data into a company's own systems, App Review status that determines whether a permission request will be approved, and compliance status that determines whether an app retains access at all.

That distinction is precisely why the announcement is relevant to marketing technologists rather than only to software engineers. Many agencies and in-house marketing teams maintain custom Meta integrations - conversion tracking pipelines, catalog synchronization tools, webhook-driven reporting systems - that were built by developers who may not check the Meta for Developers dashboard as a matter of daily routine. A compliance lapse on one of those integrations, triggered by a missed Data Access Renewal deadline, can silently break a tracking pipeline or catalog feed without an obvious symptom until data stops flowing. Surfacing that status inside an AI coding agent, as the devtools_compliance tool is designed to do, changes where that information first becomes visible - from an email inbox that may go unchecked to a development environment a technical team is already working in.

The broader pattern the launch reflects is one PPC Land has tracked closely since Google's Ads API team first floated the idea of an MCP server in July 2025: major advertising and platform companies are steadily converting dashboard-based operations into agent-queryable tools, and the split between what those tools can only read versus what they can also write is becoming the operative distinction for anyone assessing risk in an agentic workflow. Meta's own posture across its two current MCP surfaces - broad write access on the ads side, narrow write access limited to webhooks on the developer side - suggests the company is calibrating exposure differently depending on whether the audience is spending money or shipping code.

Timeline

  • November 2024: Anthropic introduces the Model Context Protocol and later donates it to the Linux Foundation.
  • October 2024: Meta begins phased rollout of its consolidated Data Access Renewal compliance system for developers.
  • April 29, 2026: Meta launches Ads AI Connectors and a companion Ads CLI, opening its advertising system to Claude and ChatGPT with write access to campaign creation.
  • June 12, 2026: The Meta Developer Tools MCP documentation page is last updated, according to its own page metadata, ahead of the public announcement.
  • June 30, 2026: Meta publishes the announcement blog post for Developer Tools MCP, written by Zoë Lieberman, confirming the beta launch and setup instructions for Claude, Cursor, Codex, and ChatGPT.

Summary

Who: Meta Platforms, through its Meta for Developers division, announced the tool in a blog post authored by Zoë Lieberman. The system is built for developers, agencies, and technical teams that build and maintain apps on Meta's developer platform, and it is compatible with four AI clients: Claude Code, Claude Desktop, the OpenAI Codex App, ChatGPT, and Cursor.

What: Developer Tools MCP is a beta remote server, reachable at https://mcp.facebook.com/devtools, exposing 10 tools that let a connected AI agent search developer documentation, inspect app settings and security configuration, check App Review and compliance status, monitor API usage and rate limits, and list, subscribe to, and test webhook subscriptions. Write access across the entire tool set is limited to one function: creating, updating, and deleting webhook subscriptions, which requires a separate Manage scope beyond the default Read scope.

When: Meta published the announcement blog post today, June 30, 2026. The associated technical documentation page carries a last-updated date of June 12, 2026.

Where: The server operates globally wherever a developer's Meta developer account and supported MCP client are available, though Meta's own documentation states the beta rollout is gradual and access is not yet available to every developer.

Why: Meta positions the tool as a way to remove the context-switching cost of building on its platform - searching documentation, checking a separate dashboard for required actions, and diagnosing failed API calls - by surfacing that information directly inside a developer's existing coding environment. The compliance-specific tool is framed around a concrete failure mode: developers who lose app access after missing an email about a Data Access Renewal deadline, a risk the tool is designed to flag proactively before access is lost.