Prowl Plugin
Real-world market data for your agents. One MCP endpoint, 444 market data tools.
Prowl is a single MCP endpoint that gives any AI agent (Claude Code, Cursor, Codex or your own stack) 444 market data tools across 17 providers: search (102 SearchAPI engines), ads, SEO and backlinks, app stores, reviews, social, trends and finance. The agent calls what it needs through prowl_search_tools → prowl_tool_info → prowl_call_tool, each call is billed on its own from one USD wallet, and every number comes back with the tool that returned it.
One call for one number. A report when one call is not enough: prowl_analyze plans many calls, cross-checks numeric claims against independent tools and composes the result. No agent of your own? Ask Prowl Agent on prowl.chat, which runs on the same tools.
This repo is the installable plugin: a namespaced /prowl:* skill plus a ready-made MCP config. Install it and your coding agent wires up the rest itself.
Plugins included
| Plugin | Skill | What it covers |
|---|
prowl | /prowl | Install and use the Prowl MCP: token setup, core tools (prowl_list_tools, prowl_search_tools, prowl_tool_info, prowl_call_tool, prowl_analyze, async sessions, schedules), tiers, playbooks, verification, and billing. Ships the .mcp.json. |
prowl-cli | /prowl-cli | Drive the same catalog from the shell with @prowl-ai/cli: catalog commands, metered calls, analyze tiers and their cost caps, playbooks, --json output, exit codes. No MCP server required. |
Install either or both — the MCP plugin for agent tool calls, the CLI plugin for scripts, CI and terminal work.
Setup
1. Get a token (human, one time)
- Sign up at https://prowl.chat ($15 starter credit, no card).
- Optionally top up the wallet.
- Generate a
prowl_... API key under MCP Home → API keys & config (shown once).
2. Provide the token to the plugin
Either export PROWL_MCP_TOKEN, or write a token file:
mkdir -p ~/.prowl
printf '%s' "prowl_YOUR_KEY" > ~/.prowl/prowl_mcp_token
chmod 600 ~/.prowl/prowl_mcp_token
3. Install the plugin
One command, any machine with Node 18+:
npx @prowl-ai/prowl-skill
It adds the marketplace, installs both plugins, and tells you what to do next. Then:
npx @prowl-ai/prowl-skill update # refresh after a new release
npx @prowl-ai/prowl-skill statusline # wire the status line (backs up settings first)
npx @prowl-ai/prowl-skill status # what is installed, what is missing
Pipe the key in so it never reaches your shell history:
printf %s "prowl_YOUR_KEY" | npx @prowl-ai/prowl-skill token
Every command takes --dry-run, which prints what it would do and changes nothing.
By hand, if you prefer:
claude plugin marketplace add https://github.com/PROWL-AI/prowl-skill
claude plugin install prowl@prowl
claude plugin install prowl-cli@prowl # optional: the shell client skill
Agents with no plugin channel (Cursor, and others that read plain skills):
npx @prowl-ai/prowl-skill plain
That copies the skills and nothing else — no hooks, no status widget, no MCP server.
It refuses to run where the plugin is installed, because a plain copy of the same name
shadows a plugin and serves its frozen version forever.
Codex
codex plugin marketplace add https://github.com/PROWL-AI/prowl-skill
Then ensure [plugins."prowl@prowl"] enabled = true in ${CODEX_HOME:-$HOME/.codex}/config.toml and restart.
The bundled .mcp.json connects to https://prowl.chat/mcp via mcp-remote, reading the token from PROWL_MCP_TOKEN or ~/.prowl/prowl_mcp_token. The token never appears in a process's arguments: the launcher hands it to mcp-remote in the PROWL_MCP_AUTH environment variable and passes only the placeholder Authorization:${PROWL_MCP_AUTH}, which mcp-remote expands itself — so ps and pgrep -fl show no key.
With no key in either place the server refuses to start — it prints the two ways to supply one and exits 78 (EX_CONFIG). It deliberately does not fall through to mcp-remote's OAuth: that opens a browser and waits on a localhost callback, and an MCP server started by an agent is killed long before the consent screen comes back, leaving the operator holding a redirect into a dead port.
Manual MCP config (any client)
Configure a Streamable HTTP MCP server at https://prowl.chat/mcp with header Authorization: Bearer prowl_YOUR_KEY. See https://prowl.chat/mcp/install.md for per-client snippets (Cursor, Claude Code, Codex, generic).
Verify
curl https://prowl.chat/mcp/health
Then, from your agent, run prowl_list_tools (free) → 444 tools grouped by category.
The status widget
A prowl_analyze call runs 30 seconds to 5 minutes and an async session runs longer.
For all of that time the terminal shows a spinner, and what it cost appears only
inside a response body nobody reads. The prowl plugin ships hooks that fix both.
It calls no API. Every figure it shows was already in a response your agent
received, so the widget costs nothing and cannot spend your wallet to draw itself.
On by default, once the plugin is installed:
- a progress bar in the dock/taskbar when
prowl_session_status reports a
fraction, taken down when the session reports a terminal status
- a one-time notice when a call fails or the wallet blocks one — once per kind of
failure, so an agent retrying five times does not produce five notices
Opt-in, one edit to ~/.claude/settings.json (a plugin cannot ship a status line):
prowl · ⟳ analyze 1:12 · 12 calls 11✓ 1✗ · $0.31 · ctx 42% · ▸ call_tool✓ tool_info✓ search_tools✓
See the prowl skill for the
two commands. State lives in ~/.prowl/status/<session_id>.json beside your token, is
pruned after seven days, and the plugin never writes to your Claude Code settings.
What your agent gets
- 23 MCP tools front the whole catalog — discovery and schema lookup are free; only data calls are metered.
- Reports, when one call is not enough –
prowl_analyze plans the calls and composes up to 16 intelligence modules (8 without a subscription). A typed evidence ledger extracts claims, attacks them adversarially, then gap-fills what survives. Contradictions and refuted claims ship inside the report.
- 3 execution modes for
prowl_analyze — basic (20–100 calls, 30–90s), deep (40–300, 3–5 min, needs Exploit+) and max (60–400, 5–10 min, needs Blackops+). A key without the subscription is downgraded to basic rather than refused.
- 8 playbooks —
geo-visibility-audit, competitor-teardown, content-engine, local-and-reputation, mobile-aso, amazon-marketplace, idea-validation, channel-economics-audit.
- Artifacts and exports — infographic, PDF, PPTX, audio brief and video from
prowl_generate_artifact; Markdown and HTML from prowl_export_report. Both read the report cached against the session_id that produced it.
Prefer a terminal? Install the prowl-cli plugin above, or use @prowl-ai/cli directly.
Security
prowl_... keys are billing-bearing — every metered tool call debits your USD wallet. Treat the key as a secret. Token files and .env are git-ignored; never commit a key to a public repo.
License
MIT — see LICENSE.
