@meetdewey/mcp

MCP server for Dewey — search and research your document collections from Claude, Cursor, and any MCP-compatible agent. See the full API reference for details.

Installation

Add to your claude_desktop_config.json (or equivalent MCP client config):

{
  "mcpServers": {
    "dewey": {
      "command": "/bin/sh",
      "args": ["-lc", "npx -y @meetdewey/mcp"],
      "env": {
        "DEWEY_API_KEY": "dwy_live_...",
        "DEWEY_COLLECTION_ID": "..."
      }
    }
  }
}

Note for Windows users: replace "command": "/bin/sh" and "args": ["-lc", "npx -y @meetdewey/mcp"] with "command": "npx" and "args": ["-y", "@meetdewey/mcp"] — the login-shell wrapper is only needed on macOS/Linux to pick up Homebrew and nvm PATH entries that Claude Desktop doesn't inherit.

Environment variables

| Variable | Required | Description | |---|---|---| | DEWEY_API_KEY | Yes | Your Dewey project API key | | DEWEY_COLLECTION_ID | No | Default collection ID. When set, tools that accept collection_id will fall back to this value. | | DEWEY_API_URL | No | Override the API base URL (default: https://api.meetdewey.com/v1) |

Available tools

Search & research

| Tool | Description | |---|---| | dewey_list_collections | List all collections in the project, including description and research instructions when set | | dewey_search | Hybrid semantic + keyword search over chunk content | | dewey_scan_sections | Lightweight search over section titles and summaries | | dewey_research | Agentic research query with configurable depth — returns a grounded, cited answer | | dewey_get_section | Fetch the full Markdown content of a section by ID |

Document management

| Tool | Description | |---|---| | dewey_list_documents | List documents in a collection with their processing status | | dewey_get_document_sections | List all sections in a document (table of contents with heading levels and IDs) | | dewey_get_document_markdown | Fetch the full converted Markdown content of a document | | dewey_retry_document | Retry a failed document — clears error state and re-queues processing | | dewey_delete_document | Permanently delete a document and all its derived data |

Claims & contradictions

| Tool | Description | |---|---| | dewey_list_claims | List extracted factual claims from a collection or specific document, filterable by importance (1–5) | | dewey_list_contradictions | List detected contradictions — clusters of conflicting claims with explanations and suggested resolutions | | dewey_detect_contradictions | Trigger an async contradiction detection run across all claims in a collection | | dewey_get_contradiction_run | Get the status and stats of the latest contradiction detection run (use to poll after dewey_detect_contradictions) | | dewey_resolve_contradiction | Apply a resolution instruction to a contradiction or dismiss it |

Deduplication

| Tool | Description | |---|---| | dewey_detect_duplicates | Trigger an async deduplication run — identifies near-duplicate documents by measuring shared content across chunks and marks one per cluster as canonical | | dewey_get_duplicate_run | Get the status and stats of the latest deduplication run (use to poll after dewey_detect_duplicates) | | dewey_list_duplicate_groups | List near-duplicate groups with canonical + members and coverage percentages | | dewey_promote_duplicate_canonical | Promote a different member to canonical; old canonical becomes near_duplicate | | dewey_disband_duplicate_group | Disband a group; former members rejoin retrieval as distinct documents |

Non-canonical documents are excluded from retrieval and contradiction detection. Requires enableDeduplication: true on the collection (set via dewey_update_collection or the Dewey dashboard).

Collection settings

| Tool | Description | |---|---| | dewey_get_collection_stats | Get document count, storage, section/chunk/claim counts, and processing status breakdown | | dewey_update_collection | Update collection name, description, research instructions, visibility, and feature flags | | dewey_recompute_summaries | Re-run AI section summarization across all documents (e.g. after changing the LLM model) | | dewey_recompute_captions | Re-run AI captioning for all images and tables across all documents | | dewey_delete_collection | Permanently delete a collection and all its data |

Research instructions

Collections can have natural-language instructions that are automatically injected into the research system prompt — for example, noting units, preferred sources, or how to handle missing information. They can be set in the Dewey dashboard, via the API, or directly through dewey_update_collection. When dewey_resolve_contradiction is used to apply a resolution, the suggested instruction is also appended here automatically.

Tool usage pattern

A typical read-only research workflow:

1. dewey_list_collections — discover available collections and understand their purpose
2. dewey_scan_sections — quickly explore document structure to identify relevant sections
3. dewey_get_section — load full content of a specific section
4. dewey_search — retrieve the most relevant chunks for a focused question
5. dewey_research — run a full agentic research query for questions that need multi-step reasoning

For deeper analysis and curation:

1. dewey_get_collection_stats — assess how much content has been processed
2. dewey_list_claims — surface the most important facts across documents
3. dewey_list_contradictions — identify where documents disagree
4. dewey_resolve_contradiction — apply or dismiss each contradiction, updating collection instructions automatically

License

MIT