# GroupDocs.Search MCP Server GroupDocs.Search MCP server lets AI agents like Claude, Cursor, and Copilot index a folder of documents — PDF, Word, Excel, PowerPoint, email archives and more — and search it by content, locally on your machine, then quote the matching passages. **GroupDocs.Search MCP server** lets AI agents like Claude, Cursor, and Copilot **index a folder of documents and search it by content** — PDF, Word, Excel, PowerPoint, email archives with their attachments, and 80+ formats in all — **locally on your machine**, then quote the passages that match. Nothing is uploaded, and the index stays on disk for the next question. Run it with one command: ```bash docker run --rm -i -v $(pwd)/documents:/data \ ghcr.io/groupdocs-search/search-net-mcp:latest ``` On **Linux**, add `--user "$(id -u):$(id -g)"` before the image name — the container runs as a non-root user, and without it the index cannot be written into a folder you own ([why]({{< ref "search/net/mcp/linux-installation.md" >}}#run-the-container-as-your-user)). {{< alert style="info" >}} **This product ships via Docker only.** The packed .NET tool is 280.9 MiB — the GroupDocs.Search engine is about 200 MiB with its ONNX and SkiaSharp natives — which is over nuget.org's 250 MB package limit, so this product ships **via Docker only**, and there is no `dnx` command for it. Every command in these pages reflects that. Details: [MCP server for .NET]({{< ref "search/net/mcp/_index.md" >}}). {{< /alert >}} Then ask your agent, in plain language: > Index the contracts folder, then tell me which contracts mention a termination penalty and quote the clause. ## What you can do Eight tools (full details in the [tools reference]({{< ref "search/mcp/tools-reference/_index.md" >}})): | The user asks… | The agent calls | It gets back | |---|---|---| | "Index this folder" | [`add_to_index`]({{< ref "search/mcp/tools-reference/add-to-index.md" >}}) | documents added, documents in the index | | "Which files mention X?" | [`search`]({{< ref "search/mcp/tools-reference/search.md" >}}) | matching files as JSON — relevance, occurrences, terms; `innerPath` for attachments | | "What does it say about X?" | [`get_text_fragments`]({{< ref "search/mcp/tools-reference/get-text-fragments.md" >}}) | the passages, with words of context, per file | | "Files changed — refresh" | [`update_index`]({{< ref "search/mcp/tools-reference/update-index.md" >}}) | added / updated / removed counts | | "Forget the drafts folder" | [`remove_from_index`]({{< ref "search/mcp/tools-reference/remove-from-index.md" >}}) | documents removed and left | | "What's in the index? Can it read .pst?" | [`get_index_info`]({{< ref "search/mcp/tools-reference/get-index-info.md" >}}) | counts, paths, status; the format list | | "Start a separate index for project Y" | [`create_index`]({{< ref "search/mcp/tools-reference/create-index.md" >}}) | an empty named index | | "Why only 10 results?" | [`get_license_status`]({{< ref "search/mcp/tools-reference/get-license-status.md" >}}) | the licensing mode and its limits | ## One question, three calls The pattern behind every scenario — captured from the server, file names shortened: ```text user agent Search MCP server | "Which contracts mention a termination penalty?" | |---------------->| | | |-- add_to_index ------------------------->| reads the | | {"documentsPath":"contracts", | files once, | | "indexName":"contracts"} | stores text | |<- "5 document(s) added, 5 in the index" --| | |-- search ------------------------------->| | | {"query":"termination AND penalty"} | | |<- {"documentsFound":3, "documents":[ ... | which files | | "contracts/acme-...-agreement.pdf", | | | "contracts/globex-...-agreement.docx",| | | "contracts/initech-nda.docx" ]} | | |-- get_text_fragments ------------------->| | | {"query":"\"termination penalty\""} | | |<- === contracts/acme-...-agreement.pdf ===| the passages | | ...a termination penalty of three | | | monthly fees applies... | |<- "Two contracts carry one: Acme (3 monthly fees, 12.3), | | Globex (50% of the remaining fees). Initech excludes it."| ``` Indexing is the one slow step. After it, each question costs a search and a fragments call, however large the folder — and the answer cites the file and the passage, so you can check it. ## Grounded answers without a vector database You do not need embeddings to ask questions about a folder of documents. An agent with full-text search finds the files, reads only the matching passages, and reformulates when a query misses — research on agentic retrieval has measured keyword tools in an agent loop reaching most of the answer quality of an embedding pipeline ([arXiv 2602.23368](https://arxiv.org/abs/2602.23368v1)). What you give up is matching by meaning: the agent has to try the words the documents use. What you gain is a local index with no model calls, results you can explain, and an index that stays current with one call. ```text Embedding pipeline Agentic full-text search ------------------------------------- ------------------------------------- split, embed, store in a vector store add_to_index: a local full-text index embed the question, take top-k chunks agent searches, refines, searches again paste the chunks into the prompt get_text_fragments: quoted passages re-embed when documents change update_index re-reads what changed needs an embedding model and a store no model calls, no network at runtime matches meaning; ranking is opaque matches words; every hit is explainable ``` ## The query language in one table Quote phrases. Put `content:` in front of a date range. The rest is what you would guess: | Query | Finds | |---|---| | `"termination penalty"` | the exact phrase — **unquoted words separated by spaces match nothing** | | `termination AND penalty` · `renewal OR renews` | both words / either word, anywhere in the document | | `terminat?` · `renewl` with `fuzzy: true` | one unknown character / near spellings | | `content: daterange(2026-07-01 ~~ 2026-09-30)` | dates written in the documents in that range | | `content: (100000 ~~ 999999)` | whole numbers in that range | | `filename: q3` · `^inv[0-9]+` | a word in the file name / a regular expression | Full table, with the traps: [`search`]({{< ref "search/mcp/tools-reference/search.md" >}}#query-language). ## Scenarios * [Search a folder of contracts]({{< ref "search/mcp/use-cases/search-a-folder-of-contracts.md" >}}) — which contracts say X, and where. * [Quote matching passages, not whole files]({{< ref "search/mcp/use-cases/quote-matching-passages-not-whole-files.md" >}}) — grounded answers on a small context. * [Keep an index up to date]({{< ref "search/mcp/use-cases/keep-an-index-up-to-date.md" >}}) — one index per project, refreshed in one call. * [Search the documents in a code repository]({{< ref "search/mcp/use-cases/search-project-documents-from-claude-code.md" >}}) — the PDFs and Word files grep cannot read. * [Search email archives and attachments]({{< ref "search/mcp/use-cases/search-email-archives-locally.md" >}}) — PST, OST, MSG, EML, and what is attached. * [Find documents by date and amount]({{< ref "search/mcp/use-cases/find-documents-by-date-and-amount.md" >}}) — date ranges, number ranges, file names. * [On-premise architecture]({{< ref "search/mcp/use-cases/on-premise-document-search.md" >}}) — what stays on the machine, and what travels. ## Install for your platform Installation, prerequisites, and client configuration are platform-specific; the tools and licensing model are the same everywhere. | Platform | Status | Install and setup | |---|---|---| | .NET | **Available** (Docker only) | [MCP server for .NET]({{< ref "search/net/mcp/_index.md" >}}) | | Java | Planned | [Tell us you need it](https://forum.groupdocs.com/c/search/16) | | Python | Planned | [Tell us you need it](https://forum.groupdocs.com/c/search/16) | | Node.js | Planned | [Tell us you need it](https://forum.groupdocs.com/c/search/16) | ## Supported AI clients | Client | How it connects | |---|---| | Claude Desktop | `claude_desktop_config.json` | | Claude Code | `claude mcp add` CLI | | VS Code / GitHub Copilot | user-level or workspace `mcp.json` | | Visual Studio 2022 (17.14+) | `.mcp.json` in the solution root | | Cursor | `~/.cursor/mcp.json` | | Windsurf | `~/.codeium/windsurf/mcp_config.json` | | Cline | Cline MCP settings | | Codex CLI | `codex mcp add` CLI | | JetBrains Rider | manual registration (Settings → AI Assistant → MCP) | Exact config blocks for every client: [Register in AI clients]({{< ref "search/net/mcp/install-in-ai-clients.md" >}}). ## How it works ```text +--------------+ +----------------------+ +--------------------+ | AI client | stdio | Search MCP server | reads | documents | | Claude, VS |<------>| Docker container, |<-------| /data: contracts/, | | Code, Cursor | JSON- | GroupDocs.Search | | mail/, reports/ | | | RPC | engine in-process | reads/ +--------------------+ +--------------+ | | writes +--------------------+ | |<------>| index store | +----------------------+ | /data/.groupdocs- | | search-index/| nothing listens on a port; no network at runtime +--------------------+ ``` The server uses MCP's **local stdio transport**: your AI client starts the container as a child process and talks to it over standard input/output. No inbound ports, no external endpoints, no telemetry. Unlike the other GroupDocs MCP servers it keeps **state** — the index — inside the folder you mount, so it survives restarts. What stays local and what reaches your model provider: [On-premise architecture]({{< ref "search/mcp/use-cases/on-premise-document-search.md" >}}). ## Delivery channel | | Docker (the only channel) | |---|---| | Prerequisites | Docker. No .NET SDK — the runtime is in the image | | Image | `ghcr.io/groupdocs-search/search-net-mcp` · `groupdocs/search-net-mcp` | | Architectures | linux/amd64 + linux/arm64 (Apple Silicon native) | | Size | about 380 MiB to download — pull it once, deliberately | ## Resources * [Quick start]({{< ref "search/mcp/getting-started/_index.md" >}}) · [Use cases]({{< ref "search/mcp/use-cases/_index.md" >}}) · [Supported formats]({{< ref "search/mcp/supported-formats.md" >}}) · [Troubleshooting & FAQ]({{< ref "search/mcp/troubleshooting-faq.md" >}}) * GitHub: [server source](https://github.com/groupdocs-search/GroupDocs.Search.Mcp) · [integration tests](https://github.com/groupdocs-search/GroupDocs.Search.Mcp.Tests) * [Product page](https://products.groupdocs.com/search/mcp/) · [Docker image](https://github.com/orgs/groupdocs-search/packages/container/package/search-net-mcp) · [MCP Registry](https://registry.modelcontextprotocol.io/v0/servers?search=io.github.groupdocs-search/groupdocs-search-mcp) * The library behind it: [GroupDocs.Search for .NET]({{< ref "search/net/_index.md" >}}) * Questions: [Search forum](https://forum.groupdocs.com/c/search/16)