No processing of your file contents takes place on any server. You need no data-processing agreement with gottrix to handle confidential or business documents.
Hosted in GermanyGlobal Content Delivery
Origin servers at Hetzner in Germany, delivered worldwide through a content delivery network. Your file contents are never transmitted in the process - they never leave your device.
The Model Context Protocol (MCP) connects AI assistants to local programs and remote services. Every client expects its own file with its own rules: Claude Desktop reads claude_desktop_config.json with the key mcpServers and only starts local stdio servers from it, Claude Code reads .mcp.json with type stdio, http or sse, Cursor uses .cursor/mcp.json, Visual Studio Code expects the key servers plus optional inputs in .vscode/mcp.json, and Codex CLI describes servers as TOML tables [mcp_servers.name] in config.toml. A wrong root key or a forgotten field usually does not produce a clear error, the server simply goes missing.
The tool checks your file against a versioned profile of the chosen client (documented form as of 2025-10) or detects the format itself. Every finding names a fixed code, the JSON Pointer and the line and column: a missing command, an unknown transport, servers instead of mcpServers, a typo with the correct field name suggested, comments or trailing commas in strict JSON, undeclared input variables or placeholders that only another client understands. It also warns about shell wrappers such as bash -c, shell characters in arguments, http without encryption and invisible or direction-changing characters.
Secrets typed into the file, such as GitHub tokens, sk- keys, Slack tokens, AWS access keys, JWTs, bearer headers or credentials in URLs, are detected and named in the report only by kind and length, never in plain text. On request a redacted copy is produced in which every detected value is replaced by the placeholder syntax of your client (${NAME}, an env variable, an input variable with a password prompt, or <SET_NAME>). Detection remains a heuristic, so check the copy yourself before you share or commit it. Commands, arguments and URLs are only ever read as text: nothing is started, no address is resolved and no server is contacted. YAML is deliberately not evaluated, and input over 1 MB or with more than 64 levels is refused before it is read.
In the Build entry mode a command line such as npx -y package-name or a URL becomes a ready server entry in the syntax of the chosen client; quotes are respected, but nothing is substituted or executed, and variable names become placeholders instead of plain-text values. The generated entry immediately goes through the same check. In the Compare mode two configurations are set side by side server by server, even across clients, while values of environment variables and headers stay hidden. Everything runs entirely locally in your browser, offline too after the first load.
Free as often as you like: no daily cap, no account, no paywall - your input stays on your device. Many online services limit free use to a few runs per day or require an account; here everything runs locally in your browser - with no daily quota and no sign-up.
Specifications
Specifications
Input formats
Text input
Output format
TXT
Batch processing
No
Processing
Locally in your browser (JavaScript)
File upload
None
In 3 steps
Choose the mode and the client profile.
Paste the configuration, command line or URL; to compare, put the second configuration in field B.
Read the report with its locations and copy the redacted copy or the generated entry.
Limitations: A passing check is not a security certificate and not a reachability test: whether a server is trustworthy, starts or responds is not checked. The profiles encode the documented file form as of 2025-10 and were not verified live against the clients; newer fields show up as warnings. Secret detection works with known formats and field names and can miss values or flag harmless ones. The redacted copy loses comments and formatting; TOML locations are exact to the line, not the column. YAML is not supported, input is limited to 1 MB. The report uses fixed technical codes with short English messages, whose meaning this page explains.
FAQ
Is my configuration uploaded?
No. Checking, building and comparing run entirely locally in your browser. Neither the file nor any token in it leaves your device, and no server is contacted.
Which clients and versions are supported?
Claude Desktop, Claude Code, Cursor, Visual Studio Code and Codex CLI, each as a profile of the documented file form as of 2025-10. The report names the profile and its date on every run.
Does "VALID" mean the server is safe?
No. VALID only means the file matches the form of the profile. Whether the started program is trustworthy or the remote server is reachable is something a syntax check cannot answer.
How are tokens and keys handled?
They are detected, named in the report only by kind and length, and replaced by placeholders such as ${GITHUB_TOKEN} in the redacted copy. Nothing is stored.
Why does VS Code not find my server?
The most common cause: .vscode/mcp.json expects the key servers, while many guides show mcpServers from other clients. The tool reports exactly this case with its location.
Can I check YAML configurations too?
No. None of the supported clients uses YAML for these files; YAML input is recognised and deliberately not evaluated, so anchors, aliases and tags are never expanded.
Is MCP Config Validator free - are there any limits?
Yes. You can use MCP Config Validator as often as you like: no daily cap, no account, no paywall. Fixed bounds: configurations up to 1 MB, 100,000 values and 64 nesting levels.