For agents · after the beta

Built for agents.

The plan: every Echo feature as an MCP tool and an API call. Agents will find it, try it without signing up and pay per call, with the cost shown first and a hard cap on every key.

Coming after the private beta. None of this is live yet: there are no API keys, MCP server, CLI, webhooks or agent payments today.

Join the beta Read the docs

Preview · not available yet

$ npx echo-mcp login✓ key saved · workspace: Your workspace · cap: $40/mo// your MCP client config{ "mcpServers": { "echo": { "command": "npx", "args": ["echo-mcp"] } } }$ curl -H "Authorization: Bearer $ECHO_API_KEY" \  https://api.echobrandtracker.com/v1/visibility?domain=example.org

MCPServer, planned

RESTAPI and webhooks, planned

freeRead calls, planned

sandboxTry without a key, planned

capOn every key, planned

90 daysBreaking-change notice, planned

Why agents pick Echo

Agents pick tools by reading descriptions, trying a call and checking the cost. Echo is built to win each step.

Found where agents look

Planned for the official MCP Registry and major directories, with an A2A agent card, llms.txt, OpenAPI spec and Markdown docs at standard addresses.

Tool descriptions written for models

Each tool says what it does, when to use it and when not to, its cost and what it returns, in under 60 words. Verb names, typed inputs.

Try before any key

A planned sandbox with realistic data on a demo brand. No signup.

Output built for context windows

Summaries first, raw answers behind a link. Trim with fields=; every response has a token_estimate.

Errors that say how to fix them

Each error has a code, a one-line cause, the fix and the call to retry.

Predictable to run

Rate limits in headers, Retry-After on every 429, idempotency keys, cursor paging and stable IDs. 90 days’ notice before breaking changes.

// how a tool describes itself to a model{  "name": "echo_visibility_get",  "description": "Visibility score (0–100) for a domain across AI engines.    Use to check how often a brand is named or cited in AI answers.    Not for web search rankings. Free. Returns score, 30-day trend,    top 3 drivers. ~400 tokens.",  "input": { "domain": "string", "engines": ["chatgpt", "aio", "perplexity", "gemini", …] },  "annotations": { "readOnlyHint": true, "costCredits": 0 }}

// an error an agent can act on402 {  "code": "cap_exceeded",  "message": "This call needs $36.00; the key has $12.40 left this month.",  "fix": "Raise the key cap, or track 2 prompts instead of 6.",  "retry": "PATCH /v1/keys/k_29f { \"cap_usd\": 80 }",  "docs": "https://docs.echobrandtracker.com/errors/cap_exceeded.md"}

Every capability, callable.

One data model with stable IDs across REST, MCP and webhooks.

Capability	REST	MCP tool
Visibility score	GET /v1/visibility	echo.visibility
Prompts × engines	GET /v1/prompts	echo.prompts.list
Get an answer, word for word	GET /v1/answers/:id	echo.answers.get
Compare two reads	GET /v1/answers/:id/diff	echo.answers.diff
Cited sources	GET /v1/sources	echo.sources.list
Sentiment by engine	GET /v1/sentiment	echo.sentiment
Track a visibility gap	POST /v1/prompts	echo.prompts.track
Pause a prompt	PATCH /v1/prompts/:id	echo.prompts.pause
Add a client site	POST /v1/sites	echo.sites.add
Send a report	POST /v1/reports	echo.reports.send
Explain a score change	webhook visibility.changed	echo.drivers

Three ways to connect

MCP server

For assistants and coding agents

Planned: about 22 tools for any MCP client. Tools that spend credits ask first.

npx echo-mcp

REST API

For custom agents and pipelines

JSON, cursor paging and an OpenAPI 3.1 spec. Answers come back word for word, with citations and model.

GET /v1/prompts?site=…&engine=perplexity

Webhooks

For agents that react

Planned: signed events when a score moves, an answer changes, a citation is lost or won, or an engine updates its model.

answer.changed  citation.lost

Skills your agent can run

Planned: six workflows as slash commands, with one install for the MCP server and every skill.

Install · after the beta

# Claude Code/plugin install echo@echo-dev# Codexcodex plugin add echo# any other agentnpx echo-mcp skills add

Planned. These packages aren’t published yet. Skills will be plain SKILL.md files you can edit.

/echo-setup

Site setup

Saves brand names, competitors, key pages and goals once, for every other skill.Saves · SITE CONTEXT

/echo-report

Client report

Writes the weekly or monthly report for one client, in your white-label template.Saves · REPORT DRAFT

/echo-changes

What changed

Answers that changed since the last read, with the diff and which moves beat the noise.Saves · NOTHING · READ ONLY

/echo-gaps

Citation gaps

Prompts where competitors are cited and you are not, and the pages engines use instead.Saves · NOTHING · READ ONLY

/echo-prompts

Prompt ideas

Turns Search Console queries and competitor gaps into prompts worth tracking.Saves · PROMPT DRAFTS

/echo-launch

Launch check

Marks a page you published and tracks which engines pick it up over 14 days.Saves · LAUNCH NOTE

Agents draft. People approve.

Agents save their work in Echo. A person approves what goes out.

1 · Reads the brief

Starts with site context

Brand names, competitors, key pages, goals and house rules, saved once per site. Every skill reads them first, free.

echo.context  free

2 · Saves its work

Drafts land in the app

Report drafts, prompts to track and launch notes go to a review queue next to their data.

echo.save_report  echo.draft_prompts

3 · Waits for a yes

Nothing goes out on its own

Nothing is sent, published or charged until someone approves it. Each key is read-only, draft or publish.

"status": "waiting_for_review"

Safe to hand to an agent

Planned to be on by default.

Spending cap on every key

A monthly dollar cap. Calls that would exceed it fail before anything is bought.

Scoped keys

Read-only, one site or full access.

Preview the cost

Add dry_run=true to see the credits, rate and new monthly total. Nothing is charged.

Safe to retry

Send an Idempotency-Key and a repeated call is never billed twice.

Cost on every response

Every response says what it spent.

Audit log

Every call by key and agent: what it read, bought or changed. Revoke a key in one click.

How agents pay

Planned payment options. Nothing can be paid or charged during the private beta.

X402 · NO ACCOUNT

Pay per call in USDC

An agent with a wallet can use Echo without signing up. Echo replies 402 with the price; the agent pays in USDC and gets the result in the same request.USDC on Base or Solana · settles in seconds

ACCOUNT KEY · PLANNED DEFAULT

Bill it to the workspace

An agent with an Echo key spends from the workspace’s monthly bill, within the key’s cap. Best for ongoing tracking.Card · invoice · monthly on the 1st

AP2 MANDATE

Spend a person has signed off

The agent presents a signed AP2 mandate: who approved it, the limit and expiry. Echo checks it before charging.Limit · expiry · scope checked before charging

PREPAID BALANCE

Top up, then draw down

Top up an agent’s balance with USDC or card. The key stops when it runs out.Top up with USDC or card · low-balance webhook

LATER

Machine Payments Protocol (MPP) and card agent tokens

Stripe and Tempo’s MPP, Visa Intelligent Commerce and Mastercard Agent Pay, once agents use them.

x402 in three stepsAsk, see the price, pay, get the answer: all in one HTTP request.No account, key or card. The 402 carries the price in credits and USDC, quoted for two minutes.

// 1 · ask for a read that spends creditsPOST /v1/reads { "prompt": "best fact checking sites", "engines": 7 }// 2 · Echo quotes the price402 Payment Required{ "credits": 7, "rate": 1.00, "amount": "7.00", "asset": "USDC",  "networks": ["base", "solana"], "pay_to": "0xEc40…", "expires_in": 120 }// 3 · agent retries with payment attachedPOST /v1/reads  X-PAYMENT: <signed payload>200 OK  answers[7] · receipt: rcpt_8QW… · tx: 0x3fa…

Pricing for agents

Planned pricing for after the beta.

Pricing

Tracking uses credits. Everything else is free.

A credit is one prompt, read once, on one engine, for one site. Only tracking spends credits, previewed and capped. Reads, site context, drafts and Search Console data are free.See pricing

POST /v1/prompts?dry_run=true

{  "prompts": 5, "engines": 7, "cadence": "daily",  "cost": {    "credits": 1050,    "rate": 0.15,    "usd_per_month": 157.50,    "account_total_after": 777.90,    "cap": 1500.00, "within_cap": true  },  "pay_with": ["account", "x402", "ap2_mandate"],  "charged": false}

Playbooks

Tested prompts for weekly agency jobs. Copy one into your agent.

EVERY MONDAY

Client check-in

Using Echo, run /echo-changes for every client site over the last 7 days. List only moves bigger than the ± band, with the engine and the answer diff. End with one action per client.

echo-changes · free

WHEN A CITATION IS LOST

Win it back

Using Echo, list citations {site} lost this month. For each, show the page the engine cites now and what it covers that ours does not. Draft the fixes. Do not publish anything.

echo.citations · echo.answers · free

ONCE A MONTH

Prompts worth tracking

Using Echo, read the site context for {site}, then run /echo-prompts on Search Console and competitor gaps. Draft at most 10 prompts, each with one line on why. Save them for my approval.

echo-prompts · free until approved

END OF MONTH

Client report

Using Echo, run /echo-report for {client} for last month in our white-label template. Lead with the score and its ±, then three wins and two gaps. Save it as a draft.

echo-report · free

Files for machines

Planned, at standard addresses, so an agent can find, understand and pay for Echo without a web page. None are published yet.

A2A agent card/.well-known/agent-card.jsonSkills, auth and pricing for agent-to-agent discovery

MCP server card/.well-known/mcp.jsonWhere to connect, the transport, and every tool with its cost

MCP Registryio.echo/echo-mcpPlanned listing in the official registry

llms.txtechobrandtracker.com/llms.txtA plain-text map of the docs; llms-full.txt has everything

Markdown docsdocs.echobrandtracker.com/*.mdEvery page as clean Markdown

OpenAPI specapi.echobrandtracker.com/openapi.jsonEvery endpoint, schema and error

Pricingechobrandtracker.com/pricing.jsonPrice steps, rates and payment methods, readable by machines

Statusstatus.echobrandtracker.com/api.jsonEngine and API health as JSON

Changelogechobrandtracker.com/changelog.rssAPI changes, with 90 days’ notice before anything breaks

Agent keys open after the private beta.Invited teams can use workspaces, roles and sites today. Nothing is charged during the beta.

Join the beta