mkt
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| MKT_CDP_URL | No | CDP endpoint (default is http://127.0.0.1:9222) | http://127.0.0.1:9222 |
| MKT_EMAIL_USER | No | Gmail SMTP user (App Password, never stored in database) | |
| MKT_LINKEDIN_AUTO | No | Set to 1 to unlock fully automatic LinkedIn (not recommended; default is semi-automatic requiring terminal confirmation) | 0 |
| MKT_EMAIL_APP_PASSWORD | No | Gmail SMTP App Password (never stored in database) |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| mkt_memoryB | 只读运营经验:从真实操作回执汇总 channel/action 模式,不暴露目标或内容,不允许自动重放。 |
| mkt_scorecardC | 只读运营记分卡:执行回执的成功/未知/失败与业务结果计数;比率是描述口径,不是渠道效果。 |
| mkt_optimizeB | 只读优化建议:按记分卡与回执模式提出下一步;requires_human=true,不自动改预算或重放。 |
| mkt_workflowB | 只读工作流视图:列表/详情/结构校验/运行记录;审批与重放只在 CLI 人工通道。 |
| mkt_orchestrateB | 只读编排:按产品/目标/任务选择最小岗位组合、交付物、交接和表达方式;不自动组队执行或发布。 |
| mkt_gtm_designB | 只读 GTM 设计:把增长任务拆成对象×目标×机制×渠道,做约束检查并给出结果契约;不执行、不发送、不预测。 |
| mkt_capabilitiesC | 只读 GTM 能力注册表:18 能力域、40 岗位族及其职责/KPI/SOP/交付物/审批映射。 |
| mkt_templatesB | 只读 GTM 交付物模板:32 个模板的版本、必填字段与 QA;可校验实例,不写工作区。 |
| mkt_work_orderB | 只读工作操作系统:WorkOrder 列表/详情/下一步与关联 SOP run、交付物、交接;不创建、不审批。 |
| mkt_metric_measureC | 只读 KPI 计算:按版本化指标字典对提供的事件确定性测量;分母为 0 返回 null,不伪造业务结果。 |
| mkt_statusB | 全局账本聚合:按 channel/campaign/result 计数。只读。 |
| mkt_channelsB | 渠道注册表:能力(六原语子集)、验证等级(PROVEN/PORTED/UNVERIFIED)、通道矩阵(每个渠道×动作的首选通道与降级链)。只读。 |
| mkt_plan_lintA | 校验营销计划 plan.json:schema 完整性、渠道能力矩阵、波次规模不超内核日预算、失败判据预注册。返回错误清单,空=通过。只读。 |
| mkt_searchB | 渠道只读搜索(客群/讨论/对标发现)。channel ∈ x|reddit|linkedin|hn|github。只读。 |
| mkt_ledger_queryC | 账本查询:按 channel/campaign/action/target/result 过滤。只读。 |
| mkt_inboxA | 只读收件箱/首响巡检:channel ∈ hn|github|email|x。github 需 repo;hn 读 submitted 新回复;email 扫 Gmail;x 扫 vendor inbox。 |
| mkt_auditB | GitHub 仓库体检(基础/社区健康/releases/贡献者集中度/AI-readiness/Scorecard)。只读。 |
| mkt_actionA | 单条写动作(与队列同一套门禁/暂停/审批/账本)。channel ∈ x|reddit|linkedin|github|hn;action ∈ publish|reply|message|engage。默认 dry-run;go=true 需已人审 plan 或有效 routine policy 且 routine=true;UNVERIFIED 渠道实跑首跑校准仅限人工 CLI 通道,MCP 不提供绕过。email 只支持队列。 |
| mkt_campaignB | Campaign 视图:计划/人审状态/队列预检/账本/事件/实验/下一步。只读。campaign 形如 <plan_id>-w1。 |
| mkt_run_queueA | 执行一个队列(全渠道:email/hn/github/reddit/linkedin/x-reply)。默认 dry-run;go=true 实跑;UNVERIFIED 渠道首跑校准仅限人工 CLI 通道;实跑须已人审 plan,或显式 routine=true 且策略匹配(无 plan 的 adhoc 仅 CLI;授权管理不对 MCP 开放)。 |
| mkt_retro_scanA | 只读扫描本地目录,统计可读文件与可提取的事实条数;不写盘、不上传。用于复盘前的体量确认。 |
| mkt_retro_analyzeC | 只读复盘分析:渠道账、浪费、冲突、以及日志里被反复提到的那几句话。口径仍由人钉。 |
| mkt_goalsB | 只读目标选项:按产品类型/阶段列出声量、使用、商业三类目标,每个带指标、判据、证明/不证明和推荐平台。 |
| mkt_channel_venuesB | 只读渠道库:具体的地方(subreddit/Slack/newsletter/会议),带人群、能否发、信号密度、竞争与风险。 |
| mkt_channel_planC | 只读渠道方案:目标→主/发射/辅助/旁听 + 经济学区间 + 容量 + 预注册实验。 |
| mkt_dealsC | 只读转化:阶段、漏斗、卡住的对象;读 DATA/deals.jsonl。 |
| mkt_deal_nextB | 只读转化:今天该跟谁、下一步动作、为什么、几天内。 |
| mkt_sprint_planC | 只读冲刺方案:目标方案→N 天并行工作流、小时级排期、每日目标、加速杠杆、容量校验与红线。 |
| mkt_goal_planB | 只读目标方案:目标+平台→平台分工、前两周动作、里程碑、止损线;不写文件、不发消息。 |
| mkt_strategyB | 只读策略推演:从产品事实包和复盘口径算出主渠道、保留清单、不做清单、冷启动五阶段、放大条件、风控映射。 |
| mkt_cadenceC | 只读落地设计:按四周切出动作表,每条带负责人、人工分钟与验收物。 |
| mkt_resultC | 只读结果:平台动作/任务结果/业务结果三层分开算,并对照判据。取不到的指标报「未测」,不当 0。 |
| mkt_reviseC | 只读回流:给出「改哪个参数、从多少到多少」并标参数级/策略级。样本不足会拒绝下结论。 |
| mkt_followup_dueC | 只读跟进:今天该跟谁、哪些已经跟丢超过 7 天。 |
| mkt_progressC | 只读进度:日报问「今天有什么该动没动」,周报问「这周比上周好还是差」。 |
| mkt_voiceB | 平台语气画像与开头骨架:这个平台怎么说话、开头怎么写、什么会被当成广告。 |
| mkt_voice_lintB | 检查一段文字在指定平台会不会被当成广告;返回问题列表,空表示可用。 |
| mkt_experienceC | 用户经验库:别人真跑过的平台做法,带出处与样本量;≥2 篇才算规律,1 篇只能当假设。 |
| mkt_playbooksB | 可用 playbook 清单:每个都带证据来源与风险提示,没有证据的会自己承认。 |
| mkt_signal_interpretationC | 只读校验宿主结构化语义候选;技术讨论、自报购买意图和不可接触对象不进入可执行机会。 |
| mkt_opportunityC | 只读需求发现:semantic_candidates 先过语义校验再匹配产品事实;不证明购买意愿。 |
| mkt_value_artifactC | 只读价值资产校验与业务渲染:教程/样例/测试/模板/对比/真实案例,action=lint|render。 |
| mkt_journeyC | 只读用户旅程:同 campaign/subject/time_window 过滤,返回漏斗、断点与修复建议。 |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 43 tools
Descriptions are unusually explicit about each tool's scope and negative space ("只读…不自动…"), which genuinely helps. However there is a dense cluster of near-synonymous planning/analysis views — mkt_goals vs mkt_goal_plan vs mkt_strategy vs mkt_channel_plan vs mkt_sprint_plan vs mkt_cadence vs mkt_gtm_design — and another of reporting views — mkt_status vs mkt_ledger_query vs mkt_scorecard vs mkt_result vs mkt_progress vs mkt_campaign — where an agent could easily pick the wrong one. mkt_memory vs mkt_experience vs mkt_playbooks is also hard to separate at selection time.
All names share the mkt_ prefix and snake_case, so surface consistency is fine. But the stem mixes nouns (mkt_goals, mkt_strategy, mkt_deals, mkt_voice, mkt_inbox, mkt_journey) with verbs (mkt_search, mkt_optimize, mkt_revise, mkt_audit, mkt_orchestrate), so you cannot infer from the name whether a tool lists, computes, or mutates.
43 tools is well past the heavy threshold for what is largely one marketing-ops workflow. Many planning/analysis views are narrow read-only lenses that could be consolidated into fewer parameterized tools.
The surface covers the full lifecycle: channel/capability registries, goal and plan design, lint/validation, gated execution (mkt_action, mkt_run_queue), measurement, retro, and learning/experience. The write side is intentionally thin (approval and first-run calibration live only in the human CLI), which is a defensible design but leaves no MCP path to author or approve plans.