MCP server

The MCP endpoint, in full

MCP (Model Context Protocol) lets an AI agent enumerate and call tools over a standard wire format. FounderReply exposes its agent at a single endpoint — POST /api/mcp — speaking JSON-RPC 2.0 over one HTTP request and one response.

Wire contract

The whole contract, transcribed from source

Endpoint: POST /api/mcp
Protocol: JSON-RPC 2.0 over one HTTP request/response
Methods: tools/list · tools/call
Auth: Authorization: Bearer fr_live_… — missing or revoked → HTTP 401
Scope: The key resolves to one workspace; params never carry a workspace id
Errors: −32700 parse · −32600 request · −32601 method · −32602 params · −32603 internal — all at HTTP 200
Not shipped: SSE / streamable-HTTP transport: server-push, notifications, sessions

QuickstartPOST /api/mcp

curl -s https://founderreply.com/api/mcp \
  -H "Authorization: Bearer fr_live_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Requesttools/call

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "set_autonomy",
    "arguments": { "action": "reply", "mode": "approve" }
  }
}

Response200 · application/json

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [
      { "type": "text",
        "text": "{\"ok\":true,\"action\":\"reply\",\"mode\":\"approve\"}" }
    ],
    "structuredContent": {
      "ok": true, "action": "reply", "mode": "approve"
    },
    "isError": false
  }
}

A key is bound to one workspace and the tools take no workspace id, so a key can only ever operate its own. And set_autonomy records exactly what you set — decide() reports any gap against a platform’s own policy at decision time, so no call arriving over this endpoint can push an action past what a platform allows.

Tool registry

Every MCP tool, and what it actually does

The whole registry — 36 tools, in the order tools/list returns them. Nothing below is a roadmap item.

FounderReply MCP tools, grouped by area, with the purpose of each tool
Tool	What it does
Persona & strategy
create_persona	Generate and persist a brand persona / content strategy from a company-focus brief, and update the workspace brand voice.
bootstrap_focus	Infer that brief from a connected account’s own recent posts instead of typing it.
get_persona	The workspace’s latest persona, or null if none exists yet.
Content
generate_content_plan	Build a rolling calendar of planned posts from the persona over the next horizonDays.
plan_news_posts	Find recent news in the brand’s field and schedule one planned post per item on the news template.
generate_blog_posts	Research SEO topics and generate finished, ready-to-publish articles in the brand voice.
list_blog_posts	Generated articles with bodies omitted, optionally filtered by status.
get_blog_post	One article by id, including its full markdown body.
list_templates	The available post formats — news, release, personal-update, tip, story, question, announcement.
list_scheduled_posts	Scheduled posts, optionally filtered by status.
Autonomy & analytics
set_autonomy	Set the mode (auto \| approve \| off) for an action type. It is applied as written; decide() reports when it exceeds a platform’s own policy.
get_analytics	Engagement overview over a trailing window, with a per-content-pillar rollup.
Growth
find_growth_opportunities	Discover public conversations the brand could add value to, and persist them.
list_opportunities	Engagement opportunities, optionally filtered by status.
Engagement — capability-gated per platform
list_messages	Inbound DMs on one platform, normalised to { id, origin, sender, content, ts, threadId }.
list_mentions	Recent comments and mentions on the brand.
list_own_posts	The brand’s own recent posts, for persona and context.
draft_reply	Draft a reply in brand voice with cross-platform memory recall. Never sends.
send_reply	Enqueue a reply to the human approval queue. Does not dispatch.
publish_content	Enqueue an original post to the human approval queue. Does not publish.
SEO
run_seo_audit	Fetch each page rendered, score it on discrete metrics plus PageSpeed Insights, and reconcile the issue register.
get_seo_overview	Site and page counts, average page score, open issues by severity, keyword coverage, average SERP position.
list_seo_issues	Open on-page issues across the workspace: error → warn → info.
track_keyword	Add or update a tracked keyword so its SERP position is sampled over time.
get_keyword_rankings	Sample current SERP positions and return the latest known position per keyword.
sync_search_console	Ingest Search Console’s authoritative impressions, clicks and average position.
add_seo_site	Register a website for SEO tracking.
generate_blog_from_seo_gaps	Find tracked keywords outside the top 10, read who outranks you, and write against the gap.
generate_conquest_posts	Mine a competitor’s surface and generate posts that capture the traffic it ranks for.
refresh_stale_blog_posts	Redraft published articles in place — same slug, honest updated_at, IndexNow ping when configured.
Browser
browse_page	Read any page as markdown by driving a real browser. The cheapest read.
extract_from_page	Pull structured fields out of a page — pricing, contact details, a product list.
screenshot_page	Capture a page as a base64 image. Rides a stored session, so it can shoot logged-in pages.
list_browser_connections	Which sites this workspace has a usable sign-in for, and which are still waiting on the user.
connect_browser_account	Mint a one-time connect_url to hand the user. Never asks anyone to type a password to the agent.
run_on_page	Drive a goal in a real browser until a success condition holds. Not a posting path — those go through approvals.