pitchmachine-mcp

pitchmachine-mcp

MCP server that lets AI agents create hyper-personalized pitch microsites, manage receivers, generate pitches, and retrieve public URLs.

Category
访问服务器

README

@pitchmachine/mcp-server

Model Context Protocol server for Pitch Machine. Build hyper-personalized pitch microsites from any AI agent — Claude Desktop, Cursor, Grok Bot, Continue, or any other MCP host.

The agent creates the receiver, generates the microsite, and hands you back a public URL. Sending is deliberately left to you (or to the agent, via its own channels) — Pitch Machine gets out of the way after the artifact is ready.


Status

v0.2.2 — the friendly-install release.

Package name is @pitchmachine/mcp-server (the earlier @pitchmachine/mcp name was retired after an unpublish incident placed it under npm's 24-hour name-reserve hold; see CHANGELOG.md). Both pitchmachine-mcp and pitchmachine-mcp-server binaries are shipped, so any older config snippets continue to work.

Auth is now a long-lived agent token you mint inside the app at Settings → Agent access. Copy once, paste into your MCP host config, done. Cookie forwarding from v0.1.x still works as a compatibility path but is no longer the recommended flow. See Roadmap for what's next.

If you were on v0.1.0: don't use it. It called the wrong receiver endpoint and sent the wrong auth header, and every first tool call would have failed. Fixed in v0.1.1; details in CHANGELOG.


What you get

Four tools:

Tool What it does
pitchmachine_create_receiver Adds a receiver (a company + one contact human). Same shape as the production form; company_name is the only required field. Set audience_mode: "b2c" and fill the B2C-only strings for personal-book receivers.
pitchmachine_generate_pitch Kicks off generation and polls until the microsite is deployed.
pitchmachine_get_pitch_url Fetches the public share URL for any pitch, past or in-progress.
pitchmachine_list_pitches Lists recent pitches with status and URL.

No send tool. On purpose. See the /agents page for the reasoning; short version: platform email deliverability is still on its warmup arc, and we'd rather hand you an artifact than a low-deliverability outbox.


Install & configure — recommended flow (v0.2.0+)

1. Mint an agent token

  1. Sign in at pitchmachine.ai.
  2. Go to Settings → Agent access.
  3. Click Mint token, name it after where it'll live (e.g. Claude on my laptop), optionally set an expiry.
  4. Copy the plaintext token. It's shown exactly once. Format: pm_agent_live_<24 base62 chars>.

Treat the token like a password. It's tied to your pitcher, not a device; revoke and re-mint from Settings if it ever leaks.

2. Point your MCP host at @pitchmachine/mcp-server

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pitchmachine": {
      "command": "npx",
      "args": ["-y", "@pitchmachine/mcp-server"],
      "env": {
        "PITCHMACHINE_API_TOKEN": "pm_agent_live_replace_with_your_token"
      }
    }
  }
}

Restart Claude Desktop. The four tools appear in the tools panel.

The Copy for Claude Desktop button in Settings → Agent access hands you this exact JSON block with the token already filled in.

Cursor

.cursor/mcp.json at your repo root (or the global equivalent):

{
  "mcpServers": {
    "pitchmachine": {
      "command": "npx",
      "args": ["-y", "@pitchmachine/mcp-server"],
      "env": {
        "PITCHMACHINE_API_TOKEN": "pm_agent_live_replace_with_your_token"
      }
    }
  }
}

Grok Bot / any other MCP host

The stdio command is the same. Point your host at npx -y @pitchmachine/mcp-server with PITCHMACHINE_API_TOKEN in the environment.

3. Verify with the smoke script

Before trusting the install in an agent workflow, run the live smoke test:

export PITCHMACHINE_API_TOKEN="pm_agent_live_replace_with_your_token"
git clone https://github.com/rcabanes-ops/pitchmachine-mcp
cd pitchmachine-mcp && npm install && npm run build && npm run smoke

This lists your recent pitches over the real API. If it prints an array, your token is good and the endpoint path is right. If it prints 401 unauthorized, the token was revoked or never existed.


Advanced / compatibility path — cookie forwarding (v0.1.x)

Still supported. Use this if you're on a Pitch Machine build that predates Settings → Agent access, or if you want a one-off check without minting a token. The cookie is valid for 14 days; re-copy after that.

<details> <summary>Cookie install steps</summary>

  1. Sign in at pitchmachine.ai.
  2. Open DevTools (F12 or Cmd+Opt+I).
  3. Application tab → Cookieshttps://pitchmachine.ai. (Firefox: Storage tab. Safari: enable Develop menu first.)
  4. Find pm_pitcher_sess. Copy the Value column.
  5. In your MCP host config, use PITCHMACHINE_SESSION_COOKIE instead of PITCHMACHINE_API_TOKEN:
{
  "mcpServers": {
    "pitchmachine": {
      "command": "npx",
      "args": ["-y", "@pitchmachine/mcp-server"],
      "env": {
        "PITCHMACHINE_SESSION_COOKIE": "v2.abcd1234...hmac_signature_here"
      }
    }
  }
}

If both env vars are set, the agent token wins.

</details>


Example agent flows

"Pitch this URL to our warm lead."

The agent calls, in order:

  1. pitchmachine_create_receiver with company_name, person_name, person_email, and any notes.
  2. pitchmachine_generate_pitch with the returned receiver_id.
  3. When the tool returns status: "deployed" and public_url, the agent drops the link into your Slack / email draft / task tracker of choice.

"What did I generate this week?"

pitchmachine_list_pitches with since set to the start of the week. Returns pitch IDs, statuses, and URLs for each.

"Resume that pitch that took forever."

pitchmachine_get_pitch_url with the pitch_id from a prior run.


Environment variables

Var Required Default Purpose
PITCHMACHINE_API_TOKEN ✅ recommended (v0.2.0+) Long-lived agent token from Settings → Agent access. Format: pm_agent_live_<24 base62>.
PITCHMACHINE_SESSION_COOKIE ✅ compatibility only Value of the pm_pitcher_sess browser cookie. Use only if you can't mint a token.
PITCHMACHINE_API_BASE https://pitchmachine.ai Point at staging or a local dev server.
PITCHMACHINE_REQUEST_TIMEOUT_MS 30000 Per-request HTTP timeout (ms).

Precedence: if both PITCHMACHINE_API_TOKEN and PITCHMACHINE_SESSION_COOKIE are set, the token wins. That lets you paste the new token into your existing config without first deleting the old cookie env var.


Development

npm install
npm run build
npm test              # vitest, 100% mocked — 42 tests
npm run inspector     # open the MCP inspector against the built server

The unit tests are network-free. The client injects a fake fetch; tool tests inject a stub client. That's what guards against regressions in the code we own, but it does not prove the client talks to a real Pitch Machine server correctly — that's what scripts/smoke.mjs is for. Every release must pass both.


Changelog

v0.2.0 (this release)

  • Added: Authorization: Bearer support as the recommended auth path. Set PITCHMACHINE_API_TOKEN to a token minted at Settings → Agent access (format: pm_agent_live_<24 base62>). Tokens are long-lived (no 14-day expiry), individually revocable, and named per-device so a leaked token can be nuked without affecting the others.
  • Docs: install steps now lead with the token flow; cookie forwarding moved to an "Advanced / compatibility" section.
  • Docs: env-var table reflects the new precedence (token beats cookie).
  • Added: 6 more tests (42 total). Token-format handling, precedence, trim, and the friendly error message are pinned.
  • Unchanged: everything else. The client, tool definitions, and wire shapes are byte-for-byte identical to v0.1.1. Cookie installs from v0.1.x keep working.

v0.1.1 (2026-08-12)

  • Fixed: receiver creation called POST /api/v2/receivers. The server exposes POST /api/receivers (no /v2/). Every first tool call in v0.1.0 404'd.
  • Fixed: auth header. v0.1.0 sent Authorization: Bearer <supabase-jwt>. The Pitch Machine server did not read that header — it authenticated via the pm_pitcher_sess HttpOnly cookie. v0.1.1 forwarded the cookie value directly. Install steps updated.
  • Fixed: PITCHMACHINE_REQUEST_TIMEOUT_MS was silently ignored (env parser wrote requestTimeoutMs, main read timeoutMs). Both fixed and a regression test pinned.
  • Added: PITCHMACHINE_SESSION_COOKIE env var. PITCHMACHINE_API_TOKEN reserved for v0.2.0.
  • Added: live smoke script at scripts/smoke.mjs. Run before trusting the install.
  • Added: 11 more tests (36 total). Auth-header shape and env precedence pinned.

v0.1.0 (2026-08-11) — do not use

Compiled, tested, published; did not actually work end-to-end because of the two bugs above. Yanked from install docs. Kept on npm as a version-history artifact.


Roadmap

  • v0.2.0 (now): agent tokens + Bearer auth. Cookie fallback preserved.
  • v0.3 (next): one-click install — .mcpb bundle or Copy-config from Settings that assembles the JSON so the reader never types it by hand.
  • v0.5: HTTP + SSE transport for hosted deployments.
  • v1.0: pitchmachine_send_pitch — once platform email deliverability is ready. Follow along at pitchmachine.ai/#/agents.

License

MIT © Pitch Machine

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选