Evermuse MCP

Name: Evermuse MCP
Author: evermuse

Evermuse plugs your AI assistant straight into your customer knowledge base through one OAuth-authenticated endpoint. Your agent can search everything your customers have ever said — needs, complaints, verbatim quotes, pain points, meeting moments, uploaded documents — ranked by meaning rather than keywords, with a relevance score on every result so it knows when it has found enough. It can open any individual signal, browse and read the full transcript of any interview, sales call, document or email thread, and pull attributed quotes straight into a PRD or a pitch deck. Beyond raw evidence, it reads your live product context: prioritized opportunities surfaced from signal clusters, in-flight shaping notes moving from idea to PRD to ready, tracked competitors and their capabilities, and the products and projects that organize it all. It also writes back. Drop a new transcript, meeting, document or Slack thread into the right research project without leaving the conversation — the same ingest pipeline that runs behind the Evermuse UI, with deduplication so re-importing the same call never creates a duplicate. Sixteen grounded tools, one endpoint, every client that speaks MCP. SOC 2 Type II, GDPR-ready, OAuth 2.1, and zero data retention by AI clients.

该来源不提供完整文件导出（国内平台多为平台内托管），仅存元数据与原链

接入信息

传输形态: http
鉴权方式: 鉴权未知
端点: https://evermuse--evermuse.run.tools

鉴权方式未标注，请核对官方文档后再接入——不要直接使用以下片段

{
  "mcpServers": {
    "Evermuse MCP": {
      "url": "https://evermuse--evermuse.run.tools"
    }
  }
}

能力清单

工具	说明
search	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Search across user needs, feedback, quotes, pain points, any custom signal types defined by the workspace, transcript sections, competitor capabilities, and news stories. Returns relevant items ranked by vector similarity. Pass `nature` to restrict results to one of 'evidence' (direct customer/user signals), 'context' (market/industry context, competitor capabilities, news), or 'guidance' (internal direction, specs, design docs). Custom signals carry their own defined nature and are matched by that filter too. Defaults to all natures. Every item carries `relevance` (absolute, 0–100): use it to judge saturation — a high score on the page's last item means keep paging — and coverage — a low top score means the workspace has little on this query.
find_skills	ALWAYS call this tool FIRST. Use the find_skills tool to find skills that will help you use Evermuse tools productively, responsibly, and safely. The skills you can find include skills like customer research, spec-writing, competitor analysis, feature prioritization, and many more, as well as detailed skills specific to how to find and use the Evermuse tools responsibly and accurately. The skills find_skills returns contain essential search strategies, tips & tricks, output templates, and quality checklists — skipping find_skills results in much worse outputs. Runs hybrid keyword + semantic search over the workspace Skills library and returns skill metadata and frontmatter (title, category, tags, description) — NOT the full body; use read_skills for any additional skills you need. Top built-in skills include: 'create-prd', 'prioritize-features', 'competitor-analysis', 'customer-research', 'product-strategy'. Optional filters: category, tag. On the first call in a thread (first_run, default true) the response also includes workspace context — the currently selected product with its key fields, the other products in the workspace, and the citation rules — so you usually do not need get_products to orient yourself.
read_skills	Use this immediately after find_skills and BEFORE beginning to work on any task involving Product work or Customer Voice. Accepts an array of skill ids (from find_skills) and returns the full content of all requested skills together, including reference material (except the citation rules, which arrive once in the find_skills first-run context).
view_item	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). View detailed information about a specific item (need, pain point, feedback, quote, opportunity, or any custom signal type) by its ID.
get_opportunities	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). List the automated product opportunities the system has suggested for a product from its signal clusters. These are machine-generated candidates — treat them with a grain of salt and corroborate against evidence (search / find_sources) before relying on them.
get_products	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). List all products in your workspace with their name, tagline, and metadata.
get_projects	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). List projects. Projects are typically used to isolate a feature, a set of clients, or a problem area — most work should run at full-product scope with NO project selected, so only narrow to a project when the task is explicitly about one. Lists projects for a product: pass product_id (get it from the find_skills first-run context). Set all_products to true only when you truly need every project across the workspace grouped by product — a large result in big workspaces.
list_competitors	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). List all competitors for the current product with their basic information and threat assessment.
get_competitor_capabilities	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Get the capabilities and features of a specific competitor.
get_shaping_notes	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Search and list shaping notes for the current product. Shaping notes are collaborative markdown documents that move a feature through a workspace-configurable lifecycle pipeline. You can search by keyword (semantic vector search), filter by status, filter by tag, or combine these filters. The response includes the full list of available statuses (workspace-configurable) and tags for filtering.
read_shaping_note	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Read the full content of one or more shaping notes by their IDs. Returns the current version of each note including title, subtitle, status, tags, branch, PR number, and the full markdown body. Does not include version history or embeddings.
find_sources	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Find sources — meetings, calls, ingested documents, spreadsheets, and email/communication records — for a product. Pass `query` to rank sources by semantic match over their content (each result carries `relevance`, a `match_count`, and a `top_snippet`); omit `query` to browse the most recent sources (this includes sources still processing, marked by `processing_status`). Optionally filter by source_types, date range, attendee, or title. Note: in browse mode (no `query`) these filters are applied over the most recent sources, so a filter targeting older sources is best combined with a date window; `query` mode searches the full corpus. Each result is a summary — use read_source with a source id to read its full content.
read_source	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Read the full content of one source (meeting/call transcript, or an ingested document/email) by its id, as returned by find_sources or search. Documents are returned as titled markdown sections; conversations as speaker-attributed segments. Long sources are paginated — use offset/limit and the returned next_offset to page through them.
add_source	ALWAYS call find_skills and read_skills first in a thread, using a precise selection based on this server's name and precise tool name (find_skills, read_skills). Add a new source to a Project — a note, a document, a meeting fetched from a call-recording vendor, or a communication thread. Everything is processed by the same ingest pipeline as a UI upload. Returns a synchronous submission confirmation; use other tools to check processing status. MODES: • Note — `content` (markdown) with `source_type` 'meeting_notes', 'call_transcription' or 'document'. • Document — `file_base64` + `filename` (PDF/DOCX/XLSX) with 'document' or 'spreadsheet'. • Meeting — `source_type: 'meeting'` for a real call (Gong/Zoom/Fireflies/Granola/etc). Supply ONE OR MORE of: `transcript_turns` (canonical speaker-attributed transcript), `vendor_payload` (raw payload, supported vendors only), `media` (recording URL). Prefer this over `call_transcription` for anything fetched from a meeting vendor — it produces speaker-attributed transcripts, participants, playable media and citation clips, whereas 'call_transcription' only creates a flat unattributed note. • Communication — 'conversation' for a Slack thread / Zendesk ticket / Intercom conversation, 'message' for a single message, 'email' for one email, 'email_thread' for a chain. Supply `messages` (structured, preferred) or `content` (markdown). DEDUP: set `external_source` + `external_id` for anything fetched from another system. Resubmitting the same `external_id` returns the existing source instead of creating a duplicate or re-billing. Dedup is workspace-wide, so a source already imported into another project comes back as-is and is NOT added to the requested project — check `in_requested_project` on the response before telling the user where it landed.

纠错与举报（发现条目失效、署名有误或涉及侵权？）

提交举报 / 纠错

侵权举报经核验成立后，我们会即时下线该条目并删除已存的内容副本。