Secure credential management for Claude Desktop and MCP servers.
🔐 Encrypted credential storage (AES-128-CBC + HMAC-SHA256, via Fernet)
🛡️ Server-side credential injection — the raw key never enters the conversation
🔄 Automatic OAuth token refresh
📡 Thin forwarding layer — rate limiting, audit logging, and usage tracking are all handled by the Clavis server
Related MCP server: GhostKey
npx @clavisagent/mcp-serverOr install globally:
npm install -g @clavisagent/mcp-serverAdd the following to your Claude Desktop configuration file (claude_desktop_config.json):
{
"mcpServers": {
"clavis": {
"command": "npx",
"args": ["-y", "@clavisagent/mcp-server"],
"env": {
"CLAVIS_API_KEY": "eyJ..."
}
}
}
}claude mcp add clavis -- npx -y @clavisagent/mcp-serverVariable | Required | Default | Description |
| yes | — | Your Clavis JWT, from |
| no |
| Base URL of your Clavis instance. Set this for self-hosted deployments. |
Tool | Description |
| Recommended. Make an API call with server-side credential injection — the credential is injected into the upstream request server-side, so the raw key never enters the conversation. |
| Legacy. Returns the raw access token or API key for a named service. Prefer |
| List all services with stored credentials |
| Check the status and expiry of credentials for a service |
Prefer call_service over get_credentials. call_service keeps the secret
server-side, so a prompt injection has no credential in context to exfiltrate.
get_credentials places the raw key in the conversation and exists only for
callers that must hold the token themselves.
MIT
RECOMMENDED: Make an API call with server-side credential injection. The credential is fetched from the Clavis vault and injected into the upstream request server-side — the raw API key never enters this conversation. Prefer this over get_credentials: it eliminates prompt-injection exfiltration as an attack vector, because there is no secret in context to exfiltrate.
| Name | Required | Description | Default |
|---|---|---|---|
| url | Yes | Full URL to call. Must be on the service connector's allowed domain — cross-domain URLs are rejected server-side (SSRF protection). | |
| data | No | Form-encoded or raw body (mutually exclusive with json) | |
| json | No | JSON request body (mutually exclusive with data) | |
| method | Yes | HTTP method for the upstream request | |
| params | No | URL query parameters | |
| headers | No | Additional request headers. Auth headers are injected server-side and cannot be overridden. | |
| service_name | Yes | Name of the Clavis service whose credentials to inject (e.g. 'my-openai') |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden of behavioral disclosure. It explains the security model: the credential is fetched from Clavis vault and injected server-side, so the raw API key never enters the conversation. It also states that this prevents exfiltration, adding meaningful context beyond the schema.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three sentences long, each serving a distinct purpose: stating the action, explaining the credential handling, and providing usage guidance. It is front-loaded with 'RECOMMENDED' and wastes no words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the tool's purpose, security behavior, and preferred usage, while the schema exhaustively documents parameters. Although there is no output schema and the description doesn't explicitly state the return format, the tool is an API caller so the response is presumably the upstream response; this is a minor gap given the overall context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema fully documents each parameter. The description reinforces the purpose of service_name (credential injection) but doesn't add new parameter-level details beyond what the schema already provides, earning the baseline score.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's function: 'Make an API call with server-side credential injection.' It names the specific resource (Clavis vault credentials) and distinguishes itself from sibling tool get_credentials, so there is no ambiguity about what this tool does.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Explicit guidance is provided: 'Prefer this over get_credentials' with a rationale that it eliminates prompt-injection exfiltration. The 'RECOMMENDED' label further signals preferred usage over the alternative.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Check the validity and rate-limit status of credentials for a service without making any external API call. Queries only the Clavis vault (database + Redis). Returns validity, expiry, rate-limit headroom, and last-used timestamp.
| Name | Required | Description | Default |
|---|---|---|---|
| service_name | Yes | The name of the service to check (e.g. 'github', 'openai') |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description fully carries the transparency burden. It discloses key behaviors: no external API call, queries only the Clavis vault (database + Redis), and specifies the returned fields (validity, expiry, rate-limit headroom, last-used timestamp). This gives a clear picture of the tool's operation and side effects.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three concise sentences, each earning its place: purpose, mechanism, and return values. It front-loads the main action and avoids any fluff or repetition.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's simplicity (1 parameter, no output schema), the description is complete. It explains what the tool does, how it works, and what it returns, covering all essential aspects without requiring an output schema.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% and the single parameter service_name is well-described in the schema. The description adds no additional meaning beyond what the schema provides, so it stays at the baseline for high coverage.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb ('Check') and a clear resource ('credentials for a service'), and explicitly states the scope ('validity and rate-limit status'). It distinguishes from siblings by noting 'without making any external API call', separating it from call_service and get_credentials.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description clearly implies this is for checking credential status before using a service, and the phrase 'without making any external API call' indicates when to use this over call_service. However, it does not explicitly name alternatives or provide when-not-to-use exclusions, so it falls short of full explicit guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
LEGACY — returns raw credential data into this conversation. Prefer call_service for security: call_service injects credentials server-side so the raw API key is never exposed to the model. Use this only when you must hold the token itself (e.g. handing it to a library that cannot be proxied). Clavis handles token refresh and rotation automatically.
| Name | Required | Description | Default |
|---|---|---|---|
| service_name | Yes | The name of the service to retrieve credentials for (e.g. 'github', 'openai') |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden. It clearly discloses that raw credentials are exposed to the model, flags itself as LEGACY, and notes automatic token refresh/rotation. This is comprehensive for a simple read tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Four sentences, each earning its place: purpose, security warning, usage restriction, and automatic refresh behavior. Front-loaded with the core function, no redundant content.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple one-parameter tool with no output schema, this is complete. It covers purpose, security context, alternatives, and lifecycle behavior. Nothing critical is missing.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% ('The name of the service to retrieve credentials for...'), so the baseline is 3. The tool description adds no additional parameter information beyond the schema, which is acceptable but not additive.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States exactly what it does: 'returns raw credential data into this conversation.' It uses a specific verb and resource, and distinguishes itself from the sibling tool call_service by noting it's the legacy/exposing variant.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides explicit when-to-use and when-not-to-use guidance: 'Prefer call_service for security' and 'Use this only when you must hold the token itself (e.g. handing it to a library that cannot be proxied).' This clearly differentiates from siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List all services configured in Clavis for the authenticated developer.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full burden. 'List' implies a read-only operation, and 'for the authenticated developer' gives context about authorization scope, but no details about side effects, permissions, rate limits, or return structure are disclosed. For a simple listing tool this is acceptable, but it is not rich in behavioral context.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence (12 words) that immediately states the action and scope. It is concise, front-loaded, and contains no filler or redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with no parameters, no output schema, and no annotations, the description provides the essential purpose and scope. It omits explicit mention of return value details, but for a simple 'list all' operation the output is self-evident. Slightly more detail (e.g., pagination or result format) could make it more complete, but it is sufficient for the tool's simplicity.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The tool has zero parameters, and the schema coverage is 100% (vacuously). Per the rubric, a 0-parameter tool gets a baseline of 4. The description does not need to explain parameters, and there is nothing missing.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses a specific verb ('List') and resource ('services configured in Clavis'), clearly distinguishing it from sibling tools like call_service, get_credentials, and check_credential_status. It also clarifies scope ('for the authenticated developer'), making the purpose explicit.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage (when you need to enumerate services) but provides no explicit guidance on when to prefer this over alternative tools, nor does it mention any exclusions or caveats. Sibling names suggest different operations, but the description itself does not address them.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.0.0call_servicecheck_credential_statusget_credentialslist_servicesScored across 4 tools
Each tool has a clearly distinct purpose: call_service proxies requests with server-side credential injection, get_credentials returns raw tokens for non-proxy use, list_services enumerates configured services, and check_credential_status inspects vault state without external calls. The overlap between call_service and get_credentials is explicitly addressed with usage guidance, so an agent can select correctly.
All tool names follow a consistent verb_noun pattern in lower_snake_case: call_service, get_credentials, list_services, check_credential_status. The verbs are distinct and clearly map to each tool's function.
With four tools, the set is tightly focused on core Clavis operations: using credentials, retrieving them, listing services, and checking status. This is the right granularity for a credential-management server—not bloated, not thin.
The tool surface covers the full lifecycle of interacting with Clavis-managed credentials: calling an upstream service, retrieving a raw credential when necessary, discovering available services, and checking status/rate limits. Missing features like credential creation or rotation are handled automatically by Clavis, so these are not gaps for the agent.
Credential broker for AI agents: scoped, revocable API access with policy enforcement and audit.
MCP-first toolbox for agents: KV storage, auth, queue, and utility tools. Free in early access.
One MCP endpoint for Claude, GPT & Gemini: 100+ tools + no-code connectors + agent workers.
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.