> ## Documentation Index > Fetch the complete documentation index at: https://docs.anythingmd.com/llms.txt > Use this file to discover all available pages before exploring further. # Troubleshooting > Fix a server that won't connect, a key that isn't picked up, uploads, large files and failed conversions. Your agent shows AnythingMD's messages in its own words. Each one says what happened and what to do next; this page covers the situations people ask about most. ## Connecting > **The agent says it has no AnythingMD tool** > > The client hasn't loaded the server yet. > > * **Claude Code:** start a new session and run `/mcp`. `anythingmd` should be listed as connected. > * **Codex:** run `codex mcp list` and check that `anythingmd` is enabled, then start a new session. > * **Cursor:** open **Cursor Settings → MCP** and switch `anythingmd` on. > * **VS Code:** run **MCP: List Servers** from the command palette and start `anythingmd`. > > Most clients only read their MCP config at startup, so restart after adding the server. > **Authentication failed, or the key isn't being sent** > > The client is connecting without a valid key. In order of likelihood: > > 1. **The client started before you set the variable.** Quit it completely and open it again. Editors launched from the Dock or Start menu may not see variables set in a shell profile; launch Cursor or VS Code from a terminal (`cursor .` or `code .`) to be sure. > 2. **The variable isn't set where the client runs.** In a new terminal, `echo $ANYTHINGMD_API_KEY` (PowerShell: `$env:ANYTHINGMD_API_KEY`) should print your key. > 3. **The config uses the wrong syntax for your client.** Copy it again from the [setup steps](/mcp/overview.md): Cursor and Windsurf use `${env:ANYTHINGMD_API_KEY}`, Kilo uses `{env:ANYTHINGMD_API_KEY}`, Codex takes the variable's name. > 4. **Kilo says "variable references are not allowed in project MCP headers".** The server is in a project's `kilo.json`. Move it to `~/.config/kilo/kilo.json`. > 5. **The key was revoked or deleted.** Create a new one under **Settings → API Keys**. > > If a key was exposed, follow the [leaked-key runbook](/leaked-keys.md). > **It works in Claude Code but not in Claude.ai, ChatGPT or the Claude desktop app** > > Those apps add servers as connectors that sign in through a browser (OAuth). The AnythingMD MCP server uses API keys, so it can't be added there yet. Use Claude Code, Codex, Cursor, VS Code, Windsurf or Kilo. ## Files and links > **How does the agent get a file from my computer?** > > It asks AnythingMD for a short-lived upload link and sends the file straight to storage, never through the chat. Agents that can run commands (Claude Code, Codex) upload it themselves. Agents that can't will give you a link: open it, pick the file, and tell the agent when it's done. > > Sandboxed environments that block outbound traffic must allow `mcp.anythingmd.com` and the upload host shown in the link. > **A link fails with “couldn't download” or “took longer than 25 seconds”** > > The link has to point at the file itself, such as a URL ending in `.pdf`. Pages that only show the file (Google Drive previews, YouTube, article pages) aren't documents. If the file's server is slow, the download stops after 25 seconds: download the file and ask the agent to convert the local copy instead. > **The file is too large** > > The limits are 100 MB per file and 1,000 pages per document. Split the document and convert the parts. > **The conversion failed and retrying doesn't help** > > The file probably can't be read: it may be password-protected or damaged. Save an unprotected copy or re-export it as a new PDF and try that. If it still fails, contact support with the job ID from the agent's message. Failed conversions are never charged. The full list of failure codes is in [Errors](/errors.md). ## Long documents > **The agent only read part of my document** > > Long documents come back in pages of about 20,000 tokens so one file doesn't fill the agent's context. The agent reads the first page and continues only when it needs a later section. Ask for the part you want ("summarize appendix B") or ask for the download link to get the whole Markdown file. > **The agent says the conversion is still running** > > Large documents can take a few minutes. The agent checks again on its own, or you can ask it to. Asking it to start over doesn't speed anything up and can convert the file twice. ## Credits and limits > **Out of credits** > > The agent's message says how many credits the conversion needs and links to billing. Buy a pack under **Settings → Billing**, then ask the agent to try again. Nothing is charged for the refused request. > **Rate limited** > > Keys are limited to 10 requests a minute until your workspace buys its first credit pack, then 60. The agent waits the time the message gives and retries once. > **The number of credits changed after the conversion** > > Before a conversion starts, credits for the estimated pages are held. You're charged for the pages actually converted and the rest is released. The agent reports the final amount. Every conversion has a receipt; see [Pricing and limits](/pricing-and-limits.md). > **How long are my files and results kept?** > > Conversion results and their download links stay available for 7 days after the conversion. After that, convert the file again if you need it. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.