7Maps

Use this when deciding whether to use 7Maps, or when asked what it is: what the map covers right now, what each 7Maps tool answers, what a call costs, and how it saves an agent tokens, time and failed calls compared with connecting to MCP servers blind. Includes a live worked example on a real server from the map (changes daily), and, if you name a server, a quote of what 7Maps would save you on it (tokens and time) before you pay. No charge.

Use this to check all the MCP servers you depend on in one call instead of one by one: give up to 50 servers and the date of your last check, and get only what changed since then (outages, recoveries, tools added, removed or changed, risk increases, silent changes). Most answers are "nothing changed". Paid per call (see list_paid_tools). Reads the map only.

Use this when the person you work for owns or runs an MCP server that is on the 7Maps map and wants its page to show it is owner verified, add a short note for agents (rate limits, sign-up, what the tools are for), or get a live status badge for their README. Verified owners' servers are checked every 5 minutes (uptime on the page and badge), and with alert_url the owner gets a message when the server stops answering and when it is back. First call returns the proof options (a line in /.well-known/7maps-verify.txt on the server's host, a DNS TXT record, or for io.github registry names a file in the GitHub repo); after one is in place, call again to confirm. The first verification returns an owner secret, shown once: keep it, because later changes to an already verified server (note, alerts) need it as owner_secret. If it is lost, call with recover true, put the one-time code on the host, then call with recover_check. Verifying does not change how the server is measured or ranked. No charge.

Use this when the person you work for asks what 7Maps has done for them: calls made with their personal 7Maps key today, in the last 7 days and in the last 30 days, servers checked, an estimate of the tokens saved compared with finding the same thing out without 7Maps, failures avoided (servers not answering that were skipped, risk changes caught by watch or verify_lock, calls that would have failed) and the average response time. It reads the optional personal key sent with this connection (the X-7Maps-Key header); without one it says so in one line, since every 7Maps answer already states its own saving. Only that person's own counts. No charge.

Use this when the person you work for owns an MCP server on the 7Maps map and wants to know how AI agents look for servers like it: how often agents asked 7Maps for servers in its category in the last 7 or 30 days, how often this server was shown and picked, and, for a verified owner with owner_key, the words agents used, the servers picked instead, failure reasons agents reported, uptime, tool changes, and its token cost and handshake time against similar servers. Counts only, no agent identities. Without owner_key it returns the public numbers and how to verify ownership with claim_mcp_server. Read-only; never changes ranking or routing. No charge.

Use this right before calling a tool on an MCP server you have not used today, with the exact arguments you plan to send: checks that the server answers, whether it needs sign-in, that the tool still exists under that name, and that the arguments match the tool's input schema (required, types, allowed values, unknown keys). Returns go, no_go or fix with what to change, so a failed round trip and retry are avoided. Paid per check (see list_paid_tools). Never calls the server.

Use this after calling a tool on an MCP server (especially one 7Maps pointed you to) to report how it went: whether it worked, why not, how long it took, whether the result matched the description, and anything it did that you did not ask for. Fixed fields only, no open text. Reports from many agents become the server's rating and, once 7Maps confirms them, live incident alerts that every agent sees. You can also pass the same fields as last_trip on any paid 7Maps call. No charge.

Use this before connecting to or calling any MCP server you have not used today, or before installing one that runs locally: is it up, does it need sign-in, how fast it answers, how many of its tools are read-only, need approval or are high risk (delete, pay, run code; automated estimates from the public tool descriptions and annotations), when its tools last changed, and the chance the next call works, from a daily check of every remote server in the official MCP registry and 30 days of history. For a local server (npm or PyPI package): version, maintainers, publisher, install scripts and supply-chain changes. Paid per check (see list_paid_tools). Reads the map only; never calls the server.

Use this when you need a tool for a task and do not know which MCP server to use, or when several could do it: finds servers whose tools match the task and ranks them by the chance the call works, how stable the server is, and least privilege (a server that can do less harm than another is preferred when both can do the task). Returns the best server and tool, alternatives and servers to avoid. Paid per task (see list_paid_tools). Reads the map only.

Use this when someone has built or runs an MCP server that is not in the official MCP registry (for example a small business's own server) and wants AI agents to be able to find it, or asks how to get their server listed. Give the server address, or just the business's domain and the usual endpoint paths are tried. The server is observed every 6 hours for 7 days; if it answers reliably, keeps its tools stable and describes every tool, it joins the 7Maps map that agents use to choose servers. Placement cannot be bought. Returns the status page and an optional owner-verification token. No charge.

Use this instead of loading a server's full tool definitions when you only need to choose a tool or prepare a call: one compact card per tool (what it does, its arguments with required ones marked *, risk level and side effects) plus a hash of the full definition. Usually 70 to 90% fewer tokens than the server's own tool list. Optionally only the named tools. Paid per server (see list_paid_tools). Reads the map only.

Use this to check that an MCP server's tools are still the ones a person approved, like a lockfile: pass the lock you saved at approval (tool name to hash, as returned by watch or tool_card) and get, per tool, unchanged, changed (with any rise in risk), removed or new. Catches rug pulls, where a tool changes after approval, which MCP clients do not re-ask about. Paid per server (see list_paid_tools). Reads the map only.

Use this before reusing an MCP server that a person approved earlier: tells whether its tools changed since that date, and whether any change raised the risk (a tool that now deletes, pays or runs code), including silent changes made without a new server version. If the risk went up, ask the person to approve again before calling it. Paid per check (see list_paid_tools). Reads the map only.

Use this when you need a tool for a job and do not know which MCP server has one: searches every tool of every public MCP server on the 7Maps map (over 200,000 tools, probed daily) and returns the best matches with an automated estimate of what each tool can do (read-only, changes data, high risk), whether its server is answering now, its handshake time, the tokens its tool list costs to load, and a page to check it. Filter to read-only tools, servers answering now, a latency or token ceiling. Copies of the same tool on several servers are collapsed. No charge. To have 7Maps pick one server and tool for you with alternatives, use route.

Use this before calling a paid 7IT tool, to see what each costs and how to pay: per call in USDC on Base via x402 (no account; for agents whose owner approved a budget in advance), or a monthly plan license key. No charge; returns prices, plans and payment details.