Skip to content

MCP server

Your coding agent searches your documents and cites the page.

Connect Claude Code, Codex, Cursor, VS Code or any other MCP client so your agent can search your projects and get answers that cite the page. Add one URL to your client and sign in with your DocSlurp account. There's nothing to install.

Endpoint
/v1/mcp
Transport
Streamable HTTP
Auth
OAuth or API key
Tools
Search, chat and more

Set upClaude CodeCodexCursorVS CodeOpenCodeClaude

On this page

Connect your client

Server URL

https://docslurp.io/v1/mcp

Set up Claude3 steps

  1. In Claude, open Customize → Connectors → Add custom connector. On a team, an organization owner may need to add it first.
  2. Name it DocSlurp and paste this URL. Choose browser sign-in and “Register automatically” for the OAuth client, then connect.
  3. Enable the connector for your conversation. The remote connector also works in Claude Desktop.
urlRemote MCP server URL
https://docslurp.io/v1/mcp

Claude connector guide ↗Verified October 2026

Streamable HTTPOAuth with PKCEBearer API keys

Sign in or use a key

OAuth sign-in Recommended

Add the server URL to your client and connect. Your client discovers the authorization server, opens DocSlurp in your browser, and asks you to approve access with your normal DocSlurp account.

Authorization code flow with PKCE. Your client manages the token exchange.

Server API key For unattended agents

Create a server key with a search mode and the chat scope (and ingest scope for uploads) in API keys and send it as an Authorization: Bearer header. Browser keys don't work with MCP.

Keep completed configs out of source control and revoke keys you no longer use.

Generate an API key config, or point at a self-hosted endpoint

Connect an MCP client

Let your agent search your projects through the DocSlurp MCP server. Sign in with OAuth, or use a server key.

Streamable HTTP endpoint

https://docslurp.io/v1/mcp

Add this to .vscode/mcp.json. Start the server in your client, then sign in and approve access in your browser. No API key is needed.

.vscode/mcp.json
{
  "servers": {
    "docslurp": {
      "type": "http",
      "url": "https://docslurp.io/v1/mcp"
    }
  }
}
Read the MCP setup and authentication docs

Tools

Start with listProjects, upload files, check their status, then search or chat. Read-only estimate tools (updateEstimate, searchSettingsEstimate, reprocessEstimate, fullIndexEstimate, indexingModeEstimate, graphRebuildEstimate, uploadEstimate) price an operation before you start it, and getJob reads a background job; an estimate tool is listed only to keys that could perform the operation it prices. When the deployment enables web tools, webSearch searches the public web for credits. project is optional: omit it for your organization's default project, or pass a project ID, slug or name (names match by slug, so "Legal Contracts" finds legal-contracts). An upload into a name that matches no project creates it, so pass a new name only when you explicitly want a new project. Access is checked against your organization and credential permissions on every request.

Quality and visuals: pass quality (fastest, balanced or smart) and visuals (none, figures, charts or charts-ai) on upload or extract to choose for one request, or change them for a project with proposeProjectSettings in a project chat ({"changes": {"pipeline": {"visuals": "charts"}}}). charts exports charts that carry their own data (Office charts, embedded chart specs, vector graphics) at no charge, and charts-ai also reads other charts with AI, billed per chart read. Unset visuals use the deployment's default, which getProjectSettings and the project record report as visualsDefaults with the rungs it offers. Price a choice first with uploadEstimate or reprocessEstimate, for example {"files": [{"name": "q3-earnings.pdf", "size": 2400000}], "visuals": "charts-ai"}. A rung the deployment or the project's privacy mode does not offer is refused with quality_unavailable or visuals_unavailable, and smart or charts-ai above the project's setting needs an editor's credential with the pipeline:billed-features scope. Project records also carry extractionChoices, the quality and visuals rungs your credential may request in that project, so a client can offer only those. The former chartExport tier (off, free, models) and quality words (fast, auto, good) are still accepted, deprecated. Precedence and rules are in the HTTP reference.

listProjectsDiscover

List authorized projects with slugs, the default project, document counts and processing status.

Arguments
{}

uploadUpload

Request storage targets for up to 10 files. Small files use one PUT; large files use the returned multipart part URLs and byte ranges. Upload the bytes, then follow getUploadStatus with sessionId. Completion and ingestion start automatically. Alternatively, pass urls to ingest hosted documents directly.

Arguments
{
  "files": [{ "filename": "manual.pdf", "mimeType": "application/pdf", "sizeBytes": 123456 }]
}

completeUploadLegacy fallback

Optional fallback for single-PUT sessions. Normal uploads finalize automatically and do not require this tool. Multipart completion is handled by the server reconciler or SDK.

Arguments
{
  "sessionId": "UPLOAD_SESSION_ID",
  "fileIds": ["UPLOADED_FILE_ID"]
}

getUploadStatusProgress

Pass sessionId from upload to follow automatic handoff and get project/run IDs. Then pass project and runId to check processing, searchable pages, failures and costs.

Arguments
{
  "project": "PROJECT_SLUG",
  "runId": "RUN_ID"
}

extractAuto-deleted

Works like upload, but the files and every output are deleted when the retention window ends: 48 hours free by default, longer with retentionDays (charged as storage). The user gets an emailed download link unless notifyEmail is false.

Arguments
{
  "urls": ["https://example.com/report.pdf"],
  "retentionDays": 7
}

getMediaEvidenceVideo and audio

Read an indexed video or audio document’s transcript, timestamped frame descriptions and summary. Pass the returned nextOffset to continue. Does not call AI.

Arguments
{
  "documentId": "DOCUMENT_ID"
}

suggestQueriesExplore

Read prior queries and suggestions without calling AI.

Arguments
{}

searchRetrieval

Retrieve text and citations for one q or a batch of up to 20 queries. Batch results stay in input order with per-query errors. No answer generation.

Arguments
{
  "q": "What are the support hours?"
}

chatGrounded answer

Get an answer with citations, a session ID and token usage, generated by the project’s configured model. Pass the returned sessionId to continue the chat.

Arguments
{
  "q": "What are the support hours?",
  "sessionId": "OPTIONAL_PREVIOUS_SESSION"
}

Usage and cost

Use search with q for one query or queries for a batch. Search returns portable passages and source metadata that you can send to your own model. Retrieval can incur embedding usage; chat additionally incurs generation usage. Search and parallel retrieval require search scope; chat requires chat scope. webSearch also needs search scope and is charged in web credits.

search response (excerpt)JSON
{
  "usage": {
    "inputTokens": 412,
    "outputTokens": 0,
    "totalTokens": 412,
    "costCents": 0.003,
    "currency": "USD",
    "costSource": "provider",
    "scope": "retrieval"
  }
}
  • Search responses include a usage object covering retrieval AI only.
  • Batch search includes usage per query.
  • Chat usage covers answer generation.
  • costSource is provider, estimate, mixed, none or unknown; costCents is in US cents.

Transport

The remote server speaks Streamable HTTP at /v1/mcp. Point your client at that endpoint, not at the older REST tool-call paths. Clients send ordinary HTTP requests and don't hold a long-lived connection open. Use a client version that supports Streamable HTTP and OAuth.

Stdio-only clients

If your client only supports stdio, run docslurp mcp-server from @docslurp/cli as a local bridge. After docslurp login it needs no key and refreshes its token while it runs; otherwise set DOCSLURP_API_KEY to a server key in the bridge's environment. It offers every tool except openUploader, which needs a host that renders MCP Apps. To use a local or self-hosted service, also set DOCSLURP_URL.

Prefer the remote endpoint when your client supports it. The bridge only relays requests; it isn't an interactive shell. See the CLI and stdio setup.

Configuration follows the VS Code MCP reference and Cursor MCP documentation.

Jump to