Back to skill

Security audit

didit-mcp-server

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill benign with medium confidence: the findings appear related to documented Didit authentication and deployment, while some references could not be fully checked.

The verdict is benign, not a guarantee of safety. Before installing, review the authentication and credential handling instructions, and pin the Docker image if you use that deployment example. Confidence is medium because Skillspector could not resolve several references, so evidence is incomplete.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (10)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 36)May include surrounding context.

md
, so mask them before showing them to anyone (host plus the last 4 characters, for example `hooks.slack.com/...a1B2`). On the hosted server `didit_org_create_api_key` also returns the new key's secret once, on purpose. A local server also returns an application's raw key from `didit_org_reveal_application_api_key` (local server only; on the hosted server a new key is created with `didit_org_create_api_key`) and credentials from `didit_account_verify_email` (the new applications' API keys and an access token). Large payloads (for example workflow feature configs) are summarized. A failed call returns no `structuredContent`: an error raised by a tool handler is sanitized text with a code and a message, plus field, hint and allowed values when there are any, while a refusal by the server itself (for example a write on a read-only connection) is plain text with no code.
- Open source (MIT): https://github.com/didit-protocol/mcp, published on npm as `@didit-protocol/mcp-server` (binary `did

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 56)May include surrounding context.

md
1. `POST /mcp` without a token answers `401` with `WWW-Authenticate: Bearer ... resource_metadata="https://mcp.didit.me/.well-known/oauth-protected-resource/mcp"`.
2. The client reads that metadata, then `https://business.didit.me/.well-known/oauth-authorization-server`, and registers itself with dynamic client registration if it has no client id.
3. It opens the browser for the authorization code flow with an S256 PKCE challenge; the user signs in (or registers) and approves the scopes. A public client uses PKCE with no client secret; the server metadata also lists `client_secret_post` and `client_secret_basic` for confidential clients.
4. It sends `Authorization: Bearer <access_token>` on every `POST /mcp`. Access tokens are short-lived and refreshed automatically with the refresh token.

The server advertises three scopes. Which resources each one covers is taken from the [tools reference](https://docs.didit.me/integration/mcp/tools) (checked 2026-11-18):

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

md
ll your role does not allow is refused, with the missing permission in the error. The tools reference lists the role permission each tool needs.
- Per the tools reference, `didit_session_create` needs a role with `create:sessions` or `write:sessions`. The server declares no local pre-check for it and leaves that check to the backend.
- Disconnect an app in the console under **Settings > Authentication > Connected apps**, or revoke its refresh token yourself (RFC 7009), which stops it from getting new access tokens. The hosted server remembers an accepted access token for up to 60 seconds (default `MCP_INTROSPECTION_CACHE_TTL_MS`), so a revoked access token can keep reaching the server for that long; each Didit API call a tool makes re-checks the token at the API.

The local server's `logout` revokes this way: a form-encoded `POST` to the `revocation_endpoint` that `https://business.didit.me/.well-known/oauth-authorization-server` advertises, sent by the client that obtained the token.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

md
ll your role does not allow is refused, with the missing permission in the error. The tools reference lists the role permission each tool needs.
- Per the tools reference, `didit_session_create` needs a role with `create:sessions` or `write:sessions`. The server declares no local pre-check for it and leaves that check to the backend.
- Disconnect an app in the console under **Settings > Authentication > Connected apps**, or revoke its refresh token yourself (RFC 7009), which stops it from getting new access tokens. The hosted server remembers an accepted access token for up to 60 seconds (default `MCP_INTROSPECTION_CACHE_TTL_MS`), so a revoked access token can keep reaching the server for that long; each Didit API call a tool makes re-checks the token at the API.

The local server's `logout` revokes this way: a form-encoded `POST` to the `revocation_endpoint` that `https://business.didit.me/.well-known/oauth-authorization-server` advertises, sent by the client that obtained the token.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

md
ll your role does not allow is refused, with the missing permission in the error. The tools reference lists the role permission each tool needs.
- Per the tools reference, `didit_session_create` needs a role with `create:sessions` or `write:sessions`. The server declares no local pre-check for it and leaves that check to the backend.
- Disconnect an app in the console under **Settings > Authentication > Connected apps**, or revoke its refresh token yourself (RFC 7009), which stops it from getting new access tokens. The hosted server remembers an accepted access token for up to 60 seconds (default `MCP_INTROSPECTION_CACHE_TTL_MS`), so a revoked access token can keep reaching the server for that long; each Didit API call a tool makes re-checks the token at the API.

The local server's `logout` revokes this way: a form-encoded `POST` to the `revocation_endpoint` that `https://business.didit.me/.well-known/oauth-authorization-server` advertises, sent by the client that obtained the token.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 211)May include surrounding context.

md
The terminal prints a short code and a link to `https://business.didit.me/device`. Open it on any device, sign in, check that the code matches and approve. Approve only a code you started yourself.

- Credentials are stored in `~/.config/didit/mcp-credentials.json` (override with `DIDIT_MCP_CREDENTIALS_FILE`), readable by you only. The command prints the granted scopes, never the token.
- The local server uses the stored credentials when `DIDIT_ACCESS_TOKEN` is not set, and refreshes them before they expire. Refresh tokens rotate, so do not share one credentials file between jobs that run at the same time.
- `npx -y @didit-protocol/mcp-server@5.1.67 logout` revokes the stored refresh token at the authorization server and deletes the stored credentials. The local copy is deleted even when the revocation fails (for example offline); then end the grant in the console under **Connected apps**.
- With no shell, an agent can call `didit_account_device_login_start`, show the link and code, t

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 242)May include surrounding context.

md
| Variable | Mode | Purpose |
|---|---|---|
| `DIDIT_ACCESS_TOKEN` | stdio | User Bearer token for headless runs |
| `DIDIT_MCP_CREDENTIALS_FILE` | stdio | Where `login` stores credentials (default `~/.config/didit/mcp-credentials.json`) |
| `DIDIT_TOOLSETS`, `DIDIT_READ_ONLY` | stdio | Toolsets and read-only mode for the local server |
| `DIDIT_API_BASE_URL` | both | Verification API base (default `https://verification.didit.me/v3`) |
| `DIDIT_AUTH_BASE_URL` | both | Auth API base (default `https://apx.didit.me/auth/v2`) |

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 232)May include surrounding context.

md
`claude mcp add ... -e DIDIT_ACCESS_TOKEN=<token>` puts the token in the process arguments and shell history; use the config file instead.

- `DIDIT_ACCESS_TOKEN` is a Didit **user** access token, not an application API key, and takes precedence over stored credentials. Treat it as a secret: keep it in the client's env config, never in a prompt or a repository.
- `MCP_DEFAULT_ORG` and `MCP_DEFAULT_APP` set the default organization and application for single-tenant setups; an `organization_id` or `application_id` passed in a call wins.

### Self-hosted HTTP server

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 237)May include surrounding context.

md
### Self-hosted HTTP server

Clone https://github.com/didit-protocol/mcp at a release tag, `npm install && npm run build`, then `node dist/http.js` (or `docker build -t didit-mcp .` and `docker run -p 3000:3000 --env-file .env didit-mcp`). It is an OAuth resource server and also accepts application API keys in the default `introspection` verify mode.

| Variable | Mode | Purpose |
|---|---|---|

Docker image referenced without tag or digest: 'docker run -p'.

Medium
Category
MCP Rug Pull
Confidence
75% confidence
Finding

Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.

Content

No source excerpt is available for this finding.