positioning-bridge
Enables MCP hosts to validate drafts against a company's versioned positioning pack through read-only tools for positioning context, content briefs, content checks, and explanation of findings.
README
Positioning Bridge
Positioning Bridge checks whether a draft carries the right company messages for its audience, channel, funnel stage, locale, and campaign.
a company keeps one versioned positioning pack. writers use the same policy through a CLI, MCP host, or TypeScript library. every result names the exact pillar, claim, rule, campaign, competitor, and source that caused it.
what it checks
| company decision | draft result |
|---|---|
| required positioning pillar is absent | missing_message |
| draft conflicts with an explicit policy rule | contradiction |
| claim is prohibited, unapproved, unqualified, or expired | unsupported_claim or stale_evidence |
| named competitor comparison has no registered claim | unsupported_claim |
| disclosure is missing after its trigger | required_disclosure |
| draft breaks the active campaign narrative | campaign_drift |
| optional message could make the position clearer | positioning_opportunity |
the decision is pass, needs_revision, or blocked. there is no composite brand
score. deterministic and model-assisted evidence stay labeled separately.
run the example
requirements: Node.js 22 or newer and npm.
git clone https://github.com/shashank-sn/positioning-bridge.git
cd positioning-bridge
npm ci
npm run build
node dist/cli/main.js validate --pack examples/acme/positioning.yaml
check the passing fictional draft:
node dist/cli/main.js check \
--pack examples/acme/positioning.yaml \
--content examples/acme/drafts/pass.md \
--audience platform-leader \
--channel landing-page \
--funnel-stage consideration \
--campaign campaign.launch
check from stdin and keep the structured result:
printf '%s\n' 'Acme trains on customer content.' | node dist/cli/main.js check \
--pack examples/acme/positioning.yaml \
--stdin \
--audience platform-leader \
--channel landing-page \
--funnel-stage consideration \
--json
the command exits 0 for pass, 1 for needs_revision or blocked, and 2 for a
usage or configuration error. omitted locale uses the pack's defaultLocale.
create a starter pack without overwriting an existing file:
node dist/cli/main.js init --output company-positioning.yaml
the project is open source on GitHub under the MIT license. positioning-bridge has not
been published to npm yet, so use the clone workflow above for 0.1.0.
connect an MCP host
build the repository, then give the host the absolute pack and executable paths:
{
"mcpServers": {
"positioning-bridge": {
"command": "node",
"args": [
"/absolute/path/to/positioning-bridge/dist/cli/main.js",
"serve",
"--pack",
"/absolute/path/to/company-positioning.yaml"
]
}
}
}
the server loads one pack at startup and exposes four read-only tools:
get_positioning_contextreturns applicable policy plus structured reasons;create_content_briefreturns required, approved, prohibited, and disclosure policy;check_contentreturns the decision, coverage, capabilities, and typed findings;explain_positioning_itemaccepts a policy or emitted finding ID plus context.
restart the process after changing the pack. the MCP guide contains the tool inputs, outputs, Codex command, and writer sequence.
positioning pack
a pack contains:
- context catalogs for audiences, channels, funnel stages, and locales;
- evidence sources with approval, visibility, verification, and expiry state;
- pillars and approved, review-required, or prohibited claims;
- named competitors and enforcement for unregistered comparisons;
- campaign narratives, desired actions, required messages, and exclusions;
- deterministic contradiction, disclosure, language, and campaign rules.
IDs are stable and every reference is validated before the server starts. see the pack reference, generated JSON Schema, and complete fictional Acme pack.
certainty and safety boundary
confirmed contradictions come from explicit deterministic rules. an optional
SemanticReviewer can report paraphrases and contextual conflicts. those findings stay
model_assisted, default to human review, and cannot block by themselves.
default runtime behavior:
- makes no network request;
- stores no submitted draft;
- loads only the configured pack path;
- caps pack and content size;
- exposes no policy mutation or publishing tool.
Positioning Bridge checks supplied policy and supplied evidence state. it does not prove an external fact, legal approval, customer permission, or source authenticity.
use with Hold Your Voice
Positioning Bridge and Hold Your Voice remain separate gates. Positioning Bridge checks message policy and evidence. HYV checks voice and linguistic patterns. neither package imports the other, and neither result is publication approval.
TypeScript library
import { PositioningService, loadPack } from "positioning-bridge";
const pack = await loadPack("./company-positioning.yaml");
const service = new PositioningService(pack);
const result = await service.checkContent({
content: "draft text",
context: {
audienceId: "primary-buyer",
channelId: "website",
funnelStageId: "consideration",
localeId: "en",
},
});
MCP construction is exported from positioning-bridge/mcp. the JSON Schema is exported
from positioning-bridge/schema.
development and verification
npm ci
npm run verify
npm run smoke:package
npm run verify checks formatting, linting, strict types, tests and coverage, schema
drift, architecture direction, documentation links, production build, and a live stdio
handshake. npm run smoke:package packs the exact version, checks its allowlist and
Markdown links, installs it in a clean temporary project, then checks its binary,
library export, example validation, and MCP handshake.
the component boundaries live in architecture.md and the executable architecture policy. release operations live in operations.md.
project status
0.1.0 is the local-first foundation. it does not include a hosted control plane, web
UI, remote model provider, CMS crawler, publication workflow, or performance analytics.
those boundaries are recorded in ADR 0001.
read CONTRIBUTING.md before opening a change. use GitHub Issues for reproducible bugs and bounded feature requests. use GitHub Discussions for setup and design questions. report vulnerabilities through the private route in SECURITY.md.
MIT licensed. see LICENSE.
推荐服务器
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。
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。
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。
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。
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。
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。
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。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
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。