WebMCP
Browser AI agents can use Social Fetch through WebMCP tools built into our site and dashboard: search docs, compare prices, check usage, run lookups and set up keys, with you watching.
WebMCP is a proposed web standard that lets a website hand tools to an AI agent running in your browser. Instead of reading the page and clicking buttons, the agent calls a tool such as get_pricing or run_quick_tool and gets a precise answer back.
Social Fetch registers WebMCP tools on socialfetch.dev and in the dashboard. An agent in your browser can research the API for you, then, once you're signed in, check your balance, run lookups on your credits, create API keys and set up monitors.
WebMCP or MCP?
Use the hosted MCP server when an AI client (Claude, ChatGPT, Cursor, Claude Code) should call Social Fetch on its own, from anywhere. WebMCP is for an agent working in your browser tab, with your sign-in, while you watch. Both run on the same credits.
Browser support
WebMCP is experimental. It runs in Chrome and Edge while the standard is in trial, and it may change. In other browsers the tools simply aren't there and the site works as normal.
To try it locally in Chrome, turn on chrome://flags/#enable-webmcp-testing, relaunch, and use an agent or extension that supports WebMCP (for example Chrome's Model Context Tool Inspector).
Tools on socialfetch.dev
Available to anyone, no account needed.
| Tool | What it does |
|---|---|
search_docs | Searches the docs and returns matching pages and sections. |
read_docs_page | Returns a docs page as Markdown: an outline without code samples, or the full page. |
find_api_endpoints | Finds endpoints by keyword and platform, one line each with its credit price. |
get_api_endpoint | One endpoint's parameters, SDK call and docs links. |
get_current_page | The page you're looking at, as text. |
get_pricing | Credit packs and plans with current prices in USD, GBP or EUR. |
get_mcp_setup | How to connect a named AI client to the hosted MCP server. |
start_signup | Opens sign-up. |
start_credit_purchase | Opens checkout for a credit pack. |
start_subscription | Opens checkout for a plan, billed monthly or annually. |
Tools in the dashboard
Available once you're signed in, for the workspace you're in.
| Tool | What it does |
|---|---|
get_account_overview | Workspace, role, credit balance and plan. |
get_usage_summary | Credits and requests over recent days, top endpoints. |
list_recent_requests | Your request log, filterable by errors, path, status and date. |
get_request | One request by its id, with a redacted preview of the response. |
open_request_log | Opens the request log with filters, for you to look at. |
get_billing_summary | Auto top-up, saved payment methods and recent payments. |
get_subscription | Your plan, included credits left and when it renews. |
find_api_endpoints | Endpoints by keyword and platform, with their credit price. |
get_api_endpoint | One endpoint's exact path and parameters. |
open_in_playground | Opens the Playground filled in, for you to press Send. |
call_api_endpoint | Runs one API call on your credits, like the Playground. With the Playground open, the request and response show there too. |
list_quick_tools | Quick tools (profile viewers, transcripts, calculators) with their fields and prices. |
run_quick_tool | Runs a quick tool on your credits and returns the result as Markdown. |
list_api_keys | Your API keys: names, prefixes, limits and last use. Never the secrets. |
create_api_key | Creates an API key and returns its secret once. |
list_monitor_sources | What can be monitored and the credits per check. |
list_monitors | Your monitors, their status and their spend this month. |
create_monitor | Creates a monitor and returns its cost per day and month. |
set_monitor_status | Pauses a monitor, or resumes a paused one. |
start_credit_purchase | Opens checkout for a credit pack for this workspace. |
start_subscription | Opens checkout for a plan. |
open_dashboard_page | Opens a dashboard page (usage, billing, keys, Playground…) for you to look at. |
open_docs_page | Opens a docs page. |
get_current_page | The dashboard page you're looking at, as text. |
get_mcp_setup | How to connect a named AI client to the hosted MCP server. |
Both sites also have read_result. Tool results are kept short (about 1,500 characters, as Chrome recommends), so a long result, such as a docs page or a big API response, comes back in parts. The agent reads the rest with read_result, for free, without repeating the call.
The same tools in Ask AI
The assistant in the dashboard (Ask AI) uses the same tools, with the same names and descriptions. It reads your account the same way. When you ask it to do something, like run a lookup, create a key or start a monitor, it shows a card with what will happen and what it costs, and nothing runs until you click the button. A key created from a card is shown only in the card, never to the assistant.
Safety
-
You can see what the agent does. While an agent uses the page, a small "AI agent" button sits in the bottom-left corner. Open it to see every action it took and whether it worked.
-
You stay in charge of money. The
start_*tools only open checkout. Payment happens on Stripe Checkout, whose own WebMCP tools mark paying as a step the agent must confirm with you. -
Consequential actions are marked.
create_api_keyandcreate_monitorcarry WebMCP'sconsequentialHint, so agents ask you before calling them. -
Social content is marked as untrusted. Results from
call_api_endpointandrun_quick_toolcarryuntrustedContentHintand start with a note telling the agent to treat posts, bios and comments as data, never as instructions. -
Spending has a ceiling per page. Agent tools stop after 1,000 credits on one page load and ask the agent to check with you. Reload the page to carry on.
-
Same rules as the dashboard. Every tool calls what the page itself calls, so your role's permissions, plan limits and credit metering all apply.
FAQ
Can an AI assistant buy credits for me?
Yes, two ways. An agent in your browser opens our checkout with start_credit_purchase and you pay on Stripe's page. Assistants that sell through Stripe agentic checkout, such as ChatGPT, can also sell you a pack inside the chat: see Buy in AI chat.
Does WebMCP cost extra?
No. Read-only tools are free. Lookups cost the same credits as the same call through the API, MCP or Playground, and failed calls are free.
Can an agent use my account without me?
No. The dashboard tools exist only in a tab where you are signed in, and the agent runs in your browser. Close the tab and the tools are gone.
Where is the API key secret shown?
Only in the result of create_api_key, once. Keys you already have never show their secret again. Manage keys on the API keys page.