annotations.title) and the standard MCP tool annotations (readOnlyHint,
destructiveHint, idempotentHint, openWorldHint), so your client can tell reads from writes and ask for
confirmation where appropriate.
All 27 tools work on Evermuse data: each one reads or writes only the workspace you authorized, so openWorldHint is
false for all of them. The one outbound fetch is optional: when you pass a recording URL to add_source, Evermuse
downloads that file to process it.
All tools
Clients should call
tools/list to see the exact tools available to a connection. add_signals and update_signals
appear only when the connection has the mcp:write scope, which standard OAuth connections receive by default.
create_shaping_note, update_shaping_note and the nature search filter are Labs features that are on by default;
a workspace admin can switch them off, and then they no longer appear.
How the tools fit together
The server sends instructions that guide agents through a reliable research flow:- Load the methodology.
find_skillsandread_skillsreturn Evermuse’s research guidance (how to search, weigh evidence and cite it). - Pick the product. All research data belongs to a product.
get_productslists them, andget_product_summaryexplains a product’s goals, users and features. Pass the chosenproduct_idto every product-scoped tool. - Start with a workflow tool. For a PRD, call
create_prd. For a research question, callcustomer_research, and so on. Each workflow tool returns the first round of evidence plus the methodology, next steps and quality bar for that task. - Go deeper. Use
search,view_item,find_sourcesandread_sourceto follow up on specific evidence. - Save what matters. Optionally record new sources, signals or shaping notes with the write tools.
Common conventions
Common parameters
Results
- Successful results return a text content block with JSON, and the same data as
structuredContent. Search-shaped tools open with a short prose digest (counts, date range, pagination). When guidance is attached, it comes as a separate text block. - Every citable item has a
urlthat links to the original evidence in the Evermuse app (for examplehttps://app.evermuse.com/s/<id>for a signal or/m/<id>for a meeting). Tool descriptions ask agents to cite inline using those URLs and never invent links. - Failed calls return
isError: truewith a text message such asError: Product not found: "abc" is not a product in this workspace. Call get_products …. The message explains how to recover. - Search-shaped tools trim each page to about 30,000 characters when guidance is attached, or about 80,000 without it,
and report
next_offsetso the agent can continue.
Guidance
find_skills
Find Skills. Hybrid keyword and semantic search over the workspace’s skills library: Evermuse’s built-in research and product-management methodologies plus any custom skills your team has added. Returns skill metadata (id, title, category, tags, description), not the full bodies. Free: doesn’t use credits.
Annotations:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
read_skills
Read Skills. Returns the full body and references of one or more skills. Free: doesn’t use credits.
Annotations:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Product context
get_products
Get Products. Lists every product in the workspace with its id, name, tagline, description, status, platform and segments. Call this first to choose aproduct_id. Takes no parameters except the optional workflow.
get_product_summary
Get Product Summary. Returns the latest product specification: mission and vision, problem statement, user segments, personas, and the feature inventory with statuses.get_projects
Get Projects. Lists the projects inside a product (id, name, description), or across all products.
All three tools:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Workflow tools
Workflow tools are the recommended starting point for each kind of product task. Each one runs the first step of a proven, multi-step process (a customer-evidence search, or a source lookup forsummarize_conversation). When
include_guidance is true, it also returns the full methodology for the task: next steps, quality criteria and the
output format.
Workflow tools are read-only: they create and save nothing. The agent writes the PRD, brief or analysis in the
conversation. It’s saved to Evermuse only if you separately ask for a write tool, such as create_shaping_note.
All workflow tools:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Evidence
search
Search Customer Evidence. Searches the customer evidence Evermuse has extracted from your sources: needs, feedback, quotes, pain points, Q&A, custom signal types, transcript sections, competitor capabilities and news. In semantic mode every item carries an absoluterelevance score (0–100); browse results have no score. Every item
carries a url.
It works in two modes:
- Semantic search: pass
search_query. Results are ranked by relevance. - Browse: omit
search_queryand pass at least one filter (note_types,meeting_id,date_fromordate_to). Results are newest first.
Annotations:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
view_item
View Customer Evidence Item. Returns the full details of one item fromsearch or get_opportunities, including
its source and context.
Annotations:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
get_opportunities
Get Opportunities. Lists machine-suggested product opportunities, derived from clusters of related customer signals, with evidence counts and, where available, the number of distinct companies behind each. Treat them as leads to investigate, not conclusions. Useview_item with item_type: "opportunity" for details. If a product has no stored
opportunities yet, the tool returns roadmap candidates instead (marked source: "roadmap_candidates"), which include
their problem and solution inline.
Returns
opportunities, total and next_offset (null on the last page).
Annotations: readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Sources
find_sources
Find Sources. Finds the meetings, calls, documents, spreadsheets and email or chat threads in a product. With aquery, results are ranked by semantic match and include the best-matching snippet. Without one, it lists sources
newest first, including those still processing.
Annotations:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
read_source
Read Source. Reads the content of one source. Meetings and calls return speaker-attributed transcript segments, and documents and emails return titled sections. Long sources are paged.
Returns the source name and type, participants,
segments, total_segments, has_more and next_offset.
Annotations: readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Competitors
list_competitors
List Competitors. Lists the competitors Evermuse tracks for a product, with description, website, threat level, similarity and segment.get_competitor_capabilities
Get Competitor Capabilities. Returns the tracked capabilities and announcements of one competitor, with announcement and release dates.
Both tools:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Shaping notes
Shaping notes are your team’s working documents in Evermuse: collaborative markdown notes that move through a status pipeline your workspace configures.get_shaping_notes
Get Shaping Notes. Lists shaping notes in a product, most recently updated first, without their content.
Also returns the workspace’s
available_statuses and available_tags.
read_shaping_note
Read Shaping Note. Returns the current version of one or more notes, with full markdown content and authorship.
Both tools:
readOnlyHint: true, destructiveHint: false, idempotentHint: true, openWorldHint: false.
Write tools
Five tools change data in your workspace. Every change is attributed to the signed-in user.add_source
Add Source to Project. Adds a meeting, call transcript, note, document, spreadsheet, or email or chat thread to Evermuse, through the same processing pipeline as an upload in the app. Evermuse then extracts signals from it.add_source returns a confirmation that the source was queued, not that it finished processing. Use
find_sources later to see when it’s ready.- Note:
content(markdown) withsource_typeofmeeting_notes,call_transcriptionordocument. - File:
file_base64plusfilenamewithsource_typedocument(PDF, DOCX and other office and text formats) orspreadsheet(XLSX). - Meeting:
source_type: "meeting"withtranscript_turns,mediaand/orvendor_payload. - Communication:
source_typeofconversation,message,emailoremail_thread, withmessages(preferred) orcontent.
Transcript and message text is limited to 4 MB per call.
Duplicates: for meetings and communications,
external_source plus external_id is a unique key across the
workspace. Re-submitting an existing pair returns already_exists: true with the existing source. It isn’t processed
again, so no processing credits are used (the call itself still counts as one tool call). Notes, documents and
spreadsheets aren’t de-duplicated: each call adds a new source.
add_signals
Add Signals. Records up to 100 signals (needs, feedback, quotes, pain points or your custom signal types) extracted from an existing source. Re-sending the same signal doesn’t create a duplicate: the key is source, type and content. Requiresmcp:write.
Returns the created signals and how many were
duplicates.
update_signals
Update Signals. Edits up to 100 existing signals in one atomic call. Each field you pass replaces the stored value, and signals have no version history, so agents should read the signal withview_item first. Requires
mcp:write.
create_shaping_note
Create Shaping Note. Saves a new shaping note, such as a PRD or brief the agent just wrote, to a product. The note starts its version history with this entry.update_shaping_note
Update Shaping Note. Changes an existing shaping note. Only the fields you pass change, and each one replaces the stored value. Every update adds a version-history entry, so earlier versions can be restored from the Shaping tab in Evermuse.Early-access tools
Workspaces in Evermuse’s early-access programs may see additional tools intools/list. The most visible is
Create Signal Type (create_signal_type), available to workspace owners and admins when custom signal types are
enabled. It prepares a draft custom signal type and shows it in an interactive review panel
(an MCP Apps UI resource, ui://evermuse/create-signal-type). Nothing is
created until you review the draft and click Publish to Evermuse in that panel. Drafts expire after 30 minutes.
Clients without MCP Apps support can’t publish drafts.