Claude Desktop | Dialogue API

The Dialogue MCP server exposes the corpus as tools that Claude Desktop — or any Model Context Protocol compatible client — can call directly during a conversation.

Setup

1. Get an API key

Contact ben@dialogue-ai.co to request access.

2. Configure Claude Desktop

Add the following to your Claude Desktop config file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

1 {
2   "mcpServers": {
3     "dialogue": {
4       "command": "npx",
5       "args": ["-y", "@dialogue-ai/mcp"],
6       "env": {
7         "DIALOGUE_API_KEY": "your_api_key"
8       }
9     }
10   }
11 }

Restart Claude Desktop. You will see “dialogue” appear in the tools list.

Available tools

`get_corpus_overview`

Returns aggregate statistics for the full corpus: total comment count, emotional register breakdown, life stage breakdown, top products mentioned, top behavioural signals, and top life events.

No parameters.

`search_comments`

Semantic search over the corpus using a natural language query. Returns the most relevant comments with their metadata and similarity score.

Parameter	Type	Required	Description
`query`	string	Yes	Natural language query, e.g. `"fear of redundancy affecting mortgage decisions"`
`limit`	integer	No	Number of results (1–50). Default: `10`
`threshold`	number	No	Minimum cosine similarity (0–1). Default: `0.4`
`emotional_register`	enum	No	Filter to a specific emotional tone
`life_stage`	enum	No	Filter to a specific life stage
`product`	string	No	Filter to comments mentioning a specific product

`get_quotes`

Retrieves high-engagement verbatim quotes sorted by like count. Use this to find the most resonant audience voice for a given segment.

Parameter	Type	Required	Description
`emotional_register`	enum	No	Filter by emotional tone
`life_stage`	enum	No	Filter by life stage
`life_event`	string	No	e.g. `partner_death`, `redundancy`, `first_home_purchase`
`product`	string	No	e.g. `life_insurance`, `pension_sipp`, `mortgage`
`min_likes`	integer	No	Minimum like count. Default: `0`
`limit`	integer	No	Number of results (1–50). Default: `10`

`filter_comments`

Returns a paginated filtered slice of the corpus by any combination of emotional register, life stage, product, or behavioural signal. Use for quantitative analysis or to pull raw comments for a specific segment.

Parameter	Type	Required	Description
`emotional_register`	string	No	Filter by emotional tone
`life_stage`	string	No	Filter by life stage
`product`	string	No	Filter by product mentioned
`signal`	string	No	e.g. `protection_gap_described`, `broker_distrust`, `estate_planning_intent`
`min_likes`	integer	No	Minimum like count. Default: `0`
`limit`	integer	No	Results per page (1–100). Default: `20`
`offset`	integer	No	Pagination offset. Default: `0`

`fetch_url`

Fetches a public URL and returns the readable body copy, stripped of navigation and boilerplate. Use this before analysing a website against the corpus.

Parameter	Type	Required	Description
`url`	string	Yes	Full URL to fetch, e.g. `https://example.com/life-insurance`

Example prompts

Once connected, ask Claude questions like:

“What are the most common fears UK adults have about retirement planning?”

“Find me the highest-engagement quotes from pre-retirement users worried about their pension.”

“What financial products are mentioned most often by people who have recently lost a partner?”

“Give me a breakdown of the emotional register across first-home buyers.”

“Fetch this page and analyse the copy against the corpus: https://example.com/product-page”