VidMage MCP
Connect your AI agent to VidMage MCP to discover creative tools, estimate credits, and run video, image, audio, and 3D tasks with your VidMage API key.
Prepare your VidMage MCP connection
VidMage MCP is a hosted Model Context Protocol server. Your agent connects over HTTPS; you do not need to run a VidMage server or install a VidMage npm package. Clients with native Streamable HTTP and custom headers can connect directly. The Claude Desktop option below uses a separate local bridge.
| Setting | Value |
|---|---|
| Product name | VidMage MCP |
| Server ID in these examples | vidmage |
| Server URL | https://vidmage.ai/api/mcp |
| Transport | Streamable HTTP |
| Authentication | Authorization: Bearer YOUR_API_KEY |
| Access | An active VidMage subscription and a dedicated API key |
| Credits | Generation uses your shared VidMage credit balance. Capability discovery and credit estimates consume no generation credits. |
- Sign in to the Developer Console with an active subscription and create a dedicated API key. The key is shown once; copy it directly into your client's private configuration.
- Install or update the agent you want to use, then choose its setup below. In file paths,
~means your user home directory. On Windows it usually corresponds to%USERPROFILE%; use the client's configuration command when available. - Merge the example into the existing configuration. Replace only the
vidmageserver entry if it already exists. Preserve other servers and settings. - Replace
YOUR_API_KEYlocally and keep theBearerprefix. Save the file, reload the client as described, and complete the connection check.
The display name is VidMage MCP. The short ID vidmage identifies the configured server and keeps existing setup commands compatible. This API-key flow does not use an OAuth login.
Keep completed configurations private. Do not paste your key into an agent conversation, shared repository, screenshot, or support log. On macOS and Linux, restrict credential files to your own account. The examples contain placeholders only; copying them does not activate your account.
Set up VidMage MCP in your agent
Choose your client to see its configuration file, complete example, reload steps, and status check. Each example uses the same endpoint and API key, with the field names required by that client.
Codex
1. Open your user configuration
Install or update Codex, then open ~/.codex/config.toml in a local editor. On Windows, the default path is %USERPROFILE%\.codex\config.toml. If you set CODEX_HOME, use its config.toml. Use this user-level file for VidMage MCP.
2. Add VidMage MCP
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Replace YOUR_API_KEY only in your local client configuration. Keep the Bearer prefix and do not paste the key into chat or source control.
Prefer the complete configuration shown when you create the key. Its structure is below. Replace an existing [mcp_servers.vidmage] entry instead of adding a duplicate. Preserve all other servers and settings.
[mcp_servers.vidmage]
url = "https://vidmage.ai/api/mcp"
http_headers = { Authorization = "Bearer YOUR_API_KEY" }
enabled = true
required = false
startup_timeout_sec = 30
tool_timeout_sec = 120This is a direct MCP registration. VidMage MCP uses your API key; this setup does not use codex mcp login. Keep required = false so a VidMage connection failure does not block Codex startup. Inspect the direct server error when VidMage is needed. On macOS or Linux, restrict the saved file with the command below. If you use a custom CODEX_HOME, adjust the path to that configuration file:
chmod 600 ~/.codex/config.toml3. Restart and verify
Run the command below to check the saved entry. Fully quit and restart Codex, then repeat it and open a new task. An existing task may not load newly configured tools.
codex mcp get vidmageIn a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
A saved entry or plugin card alone does not prove that VidMage MCP tools are available. If initialization fails, check the direct vidmage server error and the troubleshooting section.
Client reference: Official Codex MCP documentation.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Claude Code
1. Register the user-level server
Install or update Claude Code. In Bash or zsh, run this command as shown. On Windows, you can instead add the JSON below directly to your user configuration. It registers a placeholder in the user configuration without putting your real API key in shell history. If vidmage already exists, edit that entry instead.
claude mcp add-json vidmage '{"type":"http","url":"https://vidmage.ai/api/mcp","headers":{"Authorization":"Bearer YOUR_API_KEY"}}' --scope user2. Add your key locally
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Replace YOUR_API_KEY only in your local client configuration. Keep the Bearer prefix and do not paste the key into chat or source control.
Open ~/.claude.json in a local editor, or %USERPROFILE%\.claude.json on Windows. If CLAUDE_CONFIG_DIR is set, use the configuration location for that installation. Update only mcpServers.vidmage. Preserve every unrelated setting and server. The complete minimal JSON structure is:
{
"mcpServers": {
"vidmage": {
"type": "http",
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Keep type set to http. A URL without a transport type is not a valid Claude Code HTTP entry.
3. Restart and verify
Restart Claude Code, then run claude mcp get vidmage or claude mcp list. In the session, open /mcp and confirm the server is connected. An Added message only confirms the configuration was saved.
In a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
Client reference: Claude Code MCP reference.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Cursor
1. Open the global MCP configuration
Install or update Cursor. Open or create ~/.cursor/mcp.json, or %USERPROFILE%\.cursor\mcp.json on Windows. This global file makes VidMage MCP available across projects.
2. Add VidMage MCP and your key
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Replace YOUR_API_KEY only in your local client configuration. Keep the Bearer prefix and do not paste the key into chat or source control.
Merge the vidmage entry into the existing mcpServers object. If that entry already exists, replace it. Preserve other servers and valid JSON commas. Use the complete example below only when creating a new file.
{
"mcpServers": {
"vidmage": {
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}3. Enable, restart, and verify
Save the file and restart Cursor. Open its MCP settings or Customize panel, enable vidmage, and check its available tools. Follow any client trust or tool approval prompts.
In a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
If tools do not appear, open the Output panel and select MCP Logs. Check the URL and Authorization value. Cursor supports environment interpolation in remote headers, but this example stores the key directly in your private global configuration.
Client reference: Cursor MCP documentation.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
VS Code / GitHub Copilot
1. Open your user MCP configuration
Install or update VS Code and sign in to GitHub Copilot. In the Command Palette, run MCP: Open User Configuration. This opens the active user profile file; its location varies by operating system and profile. Use the user file for this setup.
2. Add VidMage MCP
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Merge the vidmage entry into servers and the input definition into inputs. Preserve other entries. Do not duplicate an existing input ID or server. A new file can use this complete JSON:
{
"inputs": [
{
"id": "vidmage-api-key",
"type": "promptString",
"description": "VidMage API key",
"password": true
}
],
"servers": {
"vidmage": {
"type": "http",
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer ${input:vidmage-api-key}"
}
}
}
}Save the file. VS Code will ask for the key when the server starts; paste only the key, without Bearer . The input hides the value and VS Code stores it for later use.
3. Start and verify
Run MCP: List Servers, choose vidmage, and start or restart it. Review the trust prompt. In Copilot Chat, use Agent mode and enable the VidMage MCP tools in the tool picker. Reload the VS Code window if the configuration is not picked up.
In a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
This input-based setup is for VS Code Copilot Chat. Interactive ${input:...} values are not forwarded to Agent Host sessions; use the separate Copilot CLI instructions for that client. VS Code uses the root key servers, while Copilot CLI uses mcpServers.
Client reference: VS Code MCP setup · Configuration reference.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
GitHub Copilot CLI
1. Open the user configuration
Install or update GitHub Copilot CLI and sign in. Open or create ~/.copilot/mcp-config.json, or %USERPROFILE%\.copilot\mcp-config.json on Windows. If COPILOT_HOME is set, use mcp-config.json in that directory.
2. Add VidMage MCP and your key
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Replace YOUR_API_KEY only in your local client configuration. Keep the Bearer prefix and do not paste the key into chat or source control.
Merge vidmage into the existing mcpServers object and preserve other entries. Replace an existing vidmage entry instead of adding a second copy. A new file can use:
{
"mcpServers": {
"vidmage": {
"type": "http",
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
},
"tools": [
"*"
]
}
}
}Keep tools: ["*"] to make the VidMage MCP tools available. This is the Copilot CLI schema; do not paste the VS Code servers configuration here.
3. Restart and verify
Save the file, restart Copilot CLI, and open /mcp to inspect the server. Follow the client permission prompts. If you prefer the interactive /mcp add form, choose HTTP, enter the same URL and Authorization header, set Tools to *, and save with Ctrl+S.
In a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
Client reference: GitHub Copilot CLI MCP setup.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Claude Desktop
1. Prepare the desktop bridge
Install or update Claude Desktop and install a current Node.js LTS release, including npm and npx. Confirm that these commands work in your terminal:
node --version
npx --versionThis setup uses mcp-remote, a third-party local bridge, to connect Claude Desktop to VidMage MCP with an API key. The bridge is not a VidMage package. Native remote Custom Connectors use a separate account-level flow; their documented OAuth fields do not provide this static-header configuration.
2. Open the desktop configuration
Open Claude Desktop Settings, then Developer, then Edit Config. The file is ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows.
Create a dedicated API key in the Developer Console. An active VidMage subscription is required. Replace YOUR_API_KEY only in your local client configuration. Keep the Bearer prefix and do not paste the key into chat or source control.
Merge vidmage into the existing mcpServers object. Replace an existing entry and preserve other settings. Use the example for your operating system. The key is stored in the local file; do not share the completed JSON.
3a. macOS configuration
{
"mcpServers": {
"vidmage": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://vidmage.ai/api/mcp",
"--transport",
"http-only",
"--header",
"Authorization:${VIDMAGE_AUTH_HEADER}"
],
"env": {
"VIDMAGE_AUTH_HEADER": "Bearer YOUR_API_KEY"
}
}
}
}3b. Windows configuration
The Windows example launches npx through cmd. Keep the header argument exactly as shown; the bridge expands VIDMAGE_AUTH_HEADER from its environment.
{
"mcpServers": {
"vidmage": {
"command": "cmd",
"args": [
"/c",
"npx",
"-y",
"mcp-remote",
"https://vidmage.ai/api/mcp",
"--transport",
"http-only",
"--header",
"Authorization:${VIDMAGE_AUTH_HEADER}"
],
"env": {
"VIDMAGE_AUTH_HEADER": "Bearer YOUR_API_KEY"
}
}
}
}4. Restart and verify
Save the file, fully quit Claude Desktop, and reopen it. The first launch needs network access to obtain the bridge package. Open Connectors or Developer settings and check vidmage.
In a new conversation, ask: Use VidMage MCP to call list_capabilities with a small limit. Do not generate anything. A successful tool response verifies authenticated access and consumes no credits.
If startup reports that npx was not found, check the Node.js installation and restart the app. Desktop logs are in ~/Library/Logs/Claude on macOS or %APPDATA%\Claude\logs on Windows. These instructions are based on the documented client and bridge formats; they are not a claim of end-to-end testing on every desktop version.
Client reference: Desktop configuration · Remote connector behavior · mcp-remote bridge.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Gemini CLI
Connect VidMage MCP through Gemini CLI's native Streamable HTTP support. Use the user settings file for a personal API key.
- Open
~/.gemini/settings.json. Create the file if it does not exist. This user configuration applies across projects. - Merge the
vidmageentry below into the existingmcpServersobject. Preserve other settings and server entries. Ifvidmagealready exists, update that entry instead of adding a duplicate key. - Replace
YOUR_API_KEYwith your VidMage API key. KeepBearerand the space before the key. Save this private user file and keep the key out of shared repositories. - Exit and restart Gemini CLI so it reloads the settings.
{
"mcpServers": {
"vidmage": {
"httpUrl": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Use httpUrl for Streamable HTTP. Gemini CLI reserves url for the legacy SSE transport.
Run gemini mcp list in your terminal, or /mcp inside Gemini CLI. Check that vidmage is connected and its available tools are listed. If authentication fails, check the API key and Authorization header before retrying.
Official reference: Gemini MCP setup, Gemini configuration.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Windsurf / Cascade
Use this configuration for the legacy Cascade agent in Windsurf or Devin Desktop. The current official documentation directs the newer Devin Local agent to the separate Devin CLI configuration described below.
- Open the MCPs panel in Cascade, or open Settings, then Cascade, then MCP Servers. Open the raw configuration file at
~/.codeium/windsurf/mcp_config.json. - Merge the
vidmageentry below into the existingmcpServersobject. Preserve your other server entries and settings. Update an existingvidmageentry instead of duplicating it. - Replace
YOUR_API_KEYwith your VidMage API key, keepingBearerand its trailing space. Save the file in this private user location. - Restart the application, reopen Cascade, and open its MCPs panel.
{
"mcpServers": {
"vidmage": {
"serverUrl": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Select vidmage and confirm that its tools are listed and enabled. Tool discovery is the first setup check. You can then ask Cascade to use an available read-only VidMage MCP tool after reviewing its description.
A team administrator can restrict MCP access or allowed servers. If the configuration is present but the tools remain unavailable, check your team settings as well as the API key.
This file applies to Cascade. Follow the Devin Local / Devin CLI section when using the newer Devin Local agent.
Official reference: Cascade MCP setup.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Devin Local / Devin CLI
Devin Local uses the Devin CLI MCP configuration. This is a separate setup from legacy Cascade. The following example uses an API key in an HTTP Authorization header.
- Open
~/.config/devin/mcp_config.jsonon macOS or Linux, or%APPDATA%\devin\mcp_config.jsonon Windows. Create the user configuration file if it does not exist. - Merge the
vidmageentry below into the existingmcpServersobject. Preserve other servers and settings. Update any existingvidmageentry rather than adding a duplicate. - Replace
YOUR_API_KEYwith your VidMage API key, keeping theBearerprefix and space. Save the file in this private user location. - Restart Devin CLI, or restart the desktop application before opening a new Devin Local session.
{
"mcpServers": {
"vidmage": {
"url": "https://vidmage.ai/api/mcp",
"transport": "http",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Run devin mcp list to confirm that vidmage is configured. In the agent session, ask it to list the tools available from VidMage MCP and confirm that discovery succeeds.
VidMage MCP uses the API key supplied in headers. An OAuth login or OAuth client ID is not required for this configuration.
These dedicated MCP files apply to Devin CLI v3000.3 and later. Earlier versions store mcpServers in the main config.json files; current versions migrate those entries on startup.
Official reference: Devin MCP configuration.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Cline
Cline supports VidMage MCP through Streamable HTTP in its IDE extension and CLI. Set the transport explicitly so Cline uses the correct connection type.
- In the Cline IDE panel, open the MCP Servers icon, choose Configure, then click Configure MCP Servers. This opens the settings JSON used by the extension. For Cline CLI, open
~/.cline/mcp.json. - Merge the
vidmageentry below into the existingmcpServersobject. Preserve other server entries and settings. Update an existingvidmageentry instead of creating a duplicate. - Replace
YOUR_API_KEYwith your VidMage API key. KeepBearerand the space before the key. Save the private settings file and keep it out of shared repositories. - Restart the server from Cline's MCP settings. If it does not reload, restart the IDE. For the CLI, exit and restart Cline.
{
"mcpServers": {
"vidmage": {
"type": "streamableHttp",
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
},
"disabled": false,
"autoApprove": []
}
}
}Keep type set to streamableHttp. Cline defaults to the legacy SSE transport when this field is omitted.
Check the MCP settings for the available VidMage MCP tools, then test an available read-only tool after reviewing its description. The CLI command cline mcp opens the server management wizard, where you can review enabled status and configuration.
The empty autoApprove list keeps tool calls subject to your normal approval flow. If authentication fails, check the API key and the Authorization header.
Official reference: Cline MCP setup.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
OpenCode
Add VidMage MCP as a remote server in OpenCode. OpenCode uses a top-level mcp object and supports API keys through custom HTTP headers.
- Open your user configuration at
~/.config/opencode/opencode.json. Create the file if it does not exist. - Merge the
vidmageentry below into the existingmcpobject. Preserve other configuration and server entries. Update any existingvidmageentry instead of duplicating it. - Replace
YOUR_API_KEYwith your VidMage API key, keeping theBearerprefix and space. Save this private user configuration and keep the key out of shared repositories. - Exit and restart OpenCode to load the updated configuration.
{
"mcp": {
"vidmage": {
"type": "remote",
"url": "https://vidmage.ai/api/mcp",
"oauth": false,
"enabled": true,
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Keep oauth set to false for this API key configuration. You do not need to run opencode mcp auth.
Run opencode mcp list to check server status. In an OpenCode session, ask it to list the tools available from VidMage MCP, then test an available read-only tool after reviewing its description.
OpenCode also reads opencode.json from the project root. Project settings can override user settings. If the expected configuration is not active, check for an existing project entry and keep real API keys out of committed files.
Official reference: OpenCode MCP setup, OpenCode configuration.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Kiro
Use the local Kiro IDE user configuration for this setup. Open the command palette and choose Kiro: Open user MCP config (JSON), or open ~/.kiro/settings/mcp.json.
- Merge the
vidmageentry into your existingmcpServersobject. Update an existing entry instead of adding a duplicate; preserve other servers and settings. - Replace
YOUR_API_KEYin this private file with your VidMage API key. Keep theBearerprefix and its space. - Save the file. Kiro reconnects after configuration changes. If needed, reconnect the server from the MCP panel or restart Kiro.
{
"mcpServers": {
"vidmage": {
"url": "https://vidmage.ai/api/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Confirm that vidmage is connected and enabled in the MCP panel, then run the read-only connection check in a new conversation.
Official reference: Kiro MCP configuration. This local file is for the IDE setup; other Kiro surfaces may use a different configuration entry point.
Finish by running the VidMage MCP connection check. A saved entry or tool list alone does not prove that an authenticated tool call works.
Verify the connection without generating media
- Check that your client has loaded and enabled the
vidmageserver. For Codex, fully quit and restart the application, then create a new task. - Send the prompt below in the agent. Approve the read-only tool call if the client requests it.
- Confirm that the result comes from the mounted
list_capabilitiestool and contains capability entries. The displayed tool name may include anmcp__vidmage__prefix.
Check my VidMage MCP connection. Use the mounted vidmage server to call list_capabilities with a small limit, using its current tool schema. Show the capability names returned. Do not upload files or submit a generation task. If the server or tool is unavailable, report that instead of using REST or another tool.Success means an authenticated tool call returned from VidMage MCP. A saved configuration, a plugin card, or a separate HTTP request does not establish that your agent loaded this server. This check consumes no generation credits; your agent provider may still charge for its own usage.
If the check fails, use the setup troubleshooting table before trying a generation.
Check the client capabilities you need
Direct connection requires Streamable HTTP and a custom Authorization header. A client that accepts only OAuth or unauthenticated connections needs a separate, verified integration path for this API-key service.
Claude Desktop: the documented option here uses the third-party mcp-remote bridge through a local stdio configuration. It requires Node.js and is distinct from an account-level remote connector. Follow the bridge instructions in its agent section.
ChatGPT web: this guide does not provide a verified direct API-key setup. Do not select No Authentication or paste your VidMage key into an OAuth client-secret field. Check the official ChatGPT developer-mode documentation for the authentication modes supported by your workspace.
Local media: upload_files returns upload commands, and completed tasks can return a download command. Running those commands requires local file access and command execution in the host client, or a manual terminal step. Adding MCP alone does not grant those capabilities. Use reachable media URLs where the selected schema accepts them, or complete the returned transfer manually. Never share temporary signed transfer URLs in public logs.
Other clients: use the URL, Streamable HTTP transport, and Bearer header from the connection table. Read that client's current schema instead of copying another client's JSON. These instructions document setup formats; confirm actual account access with the read-only check.
Discover only what you need
| MCP tool | Purpose |
|---|---|
list_capabilities | Find enabled operations for the output you want. |
describe_capability | Read the selected operation's current input schema. |
list_models | Find model identifiers and available choices. |
estimate_model_credits | Estimate a model request without generating media. |
estimate_capability_credits | Estimate a feature request without generating media. |
upload_files | Prepare transfers for one to five local files. |
submit_task | Submit a generation or editing task; this can use credits. |
get_task_result | Query and advance the existing authorized task, including deferred provider work and billing, then retrieve its output. |
select_faces | Apply a user's face selection to the existing task. |
get_recent_tasks | Recover recent task IDs after an interruption. |
Call list_capabilities when choosing an operation. Load the selected capability with describe_capability when you still need its input schema. When you already know the capability and current schema, skip that discovery step. For model selection and cost estimates, use list_models, followed by estimate_model_credits with the intended model settings.
Use the tool definitions exposed by the connected server for current argument shapes. Avoid copying a REST request body into an MCP tool without checking that tool's schema.
For feature tools, use estimate_capability_credits with the selected capability and its intended inputs. Both model and feature estimators are read-only. Inspect current tool definitions before building estimator arguments.
Keep the media workflow on one task
- For local files, call
upload_filesonce for one to five files in the batch, then run its returneduploadCommandvalues in parallel. - Call
submit_taskonce with a fresh, stableidempotencyKeyfor this generation request. - Call
get_task_resultfor the returned task. While it is running, wait at least the returnedretryAfterMsbefore the next call. Queries can advance provider work and billing already authorized by submission. - If the task returns
NEEDS_INPUT, pause polling and present every corrected native face preview with its stable index and callselect_facesonce with the selection. - Continue polling the same task. Supply an absolute writable
localPathwhen you need a local artifact.
The face selection branch is conditional. Do not request another selection or submit a replacement task just because the first result is still pending.
Save the completed result once
When get_task_result completes with an artifact, run its returned artifact.downloadCommand once, verify artifact.clientLocalPath exists, and present that local output. Small image, audio, or file content may also appear natively in the client. Downloading the existing artifact consumes no generation credits.
If the download or preview fails, recover that same artifact and task. A display problem does not require another generation.
Fix installation and connection problems
| Symptom or error | What to do |
|---|---|
| Server saved, but no tools appear | Enable the server, check for a malformed JSON/TOML file or duplicate vidmage entry, and reload the client. In Codex, fully restart and open a new task. Check startup logs for the direct VidMage server; a separate Plugins catalog error is unrelated. |
| Wrong transport or configuration fields | Use the matching agent example: Gemini uses httpUrl, VS Code uses the root key servers, Cline uses type: streamableHttp, and OpenCode uses mcp. Do not configure this endpoint as legacy SSE. |
AUTHENTICATION_REQUIRED | No credential reached VidMage. Replace the placeholder and check that the client sends the Authorization header. If you chose an environment-variable setup, confirm the variable is available to the running app, not only to a different terminal. |
AUTHORIZATION_HEADER_INVALID | The header is malformed. Use Authorization: Bearer YOUR_API_KEY with exactly one space between Bearer and the key, and no newline in the value. |
API_KEY_INVALID_OR_REVOKED or API_KEY_INVALID_CREDENTIAL | Create a new dedicated key in the Developer Console. Replace the credential in the existing server entry, reload the client, and repeat the read-only check. |
| Subscription or access denied | Read the returned error code and confirm that the key belongs to an account with an active subscription. Check organization restrictions on MCP servers. Do not assume every 403 means an invalid key. |
CREDENTIAL_STORAGE_UNAVAILABLE / HTTP 503 | Preserve the current key. Honor Retry-After and retry the connection once. If the problem continues, contact support with the non-secret requestId. Do not rotate keys or submit generation tasks to diagnose storage availability. |
| Timeout, 404, or proxy error | Confirm the exact HTTPS endpoint https://vidmage.ai/api/mcp. Check your network, VPN, firewall, and proxy access to vidmage.ai. Use the server's retry guidance. Keep TLS verification enabled. |
| Claude Desktop cannot start npx | Install a current Node.js LTS release and confirm that node and npx are available. Fully restart Claude Desktop. Use the Windows command wrapper shown in its section when applicable; inspect the client's MCP logs if startup still fails. |
| OAuth sign-in opens for an API-key setup | Use the static Bearer header configuration for your client. For this Codex setup, do not run codex mcp login vidmage. OpenCode should keep oauth set to false. |
| Connected, but upload or local preview fails | Check the host's file and command permissions. Finish the returned upload or download command once and verify the target path. Reuse the existing artifact and task; a display problem does not require another generation. |
For support, include the agent name and version, operating system, error code, and a redacted request ID. Keep API keys, Authorization values, signed URLs, and private media out of shared logs. See the full error reference for task and billing recovery.
Recover within the existing conversation
A disconnected client can resume with the existing task ID. If the task identity was lost, use get_recent_tasks. Preserve the original idempotency key through retries of the same submission.
For ACCOUNT_SESSION_REFRESH_REQUIRED, sign in to VidMage once. For an invalid or revoked key, replace the credential in the client. For uncertain billing or submission outcomes, follow the returned recovery action.
Update or remove the connection
To rotate a key, create its replacement in the Developer Console, update only the Authorization value in your private client configuration, reload the client, and repeat the read-only check. Revoke the old key in the console when you no longer need it.
To disconnect, disable or remove only the vidmage server entry through the client's MCP settings. For Codex CLI, codex mcp remove vidmage removes the registration. Preserve all unrelated servers and settings. Removing a local entry does not revoke the key or cancel a task already submitted to VidMage.
Choose the reference for the media task
- Seedance 2.5 parameters for video generation and media references.
- AI Photo Face Swap parameters for a target photo and replacement face.
- AI Video Face Swap parameters for a tracked face in an existing clip.
MCP and REST share the enabled capability set. Authentication, upload completion, task identity, and billing state still matter even when an AI client chooses and calls the tools.
For all other tools and models, find the exact operation in the complete reference directory.
Need help with VidMage MCP? Contact us at [email protected].
