io.github.delimit-ai/delimit-mcp-server

What we checked

Six static checks, weighted by risk. Every result reflects only what could be observed in the published package and repository — never intent.

• Injection surface

  unscannable25% of grade

  Tool descriptions/manifest scanned for instruction-injection patterns (imperatives at the model, hidden text, 'ignore previous', data-exfil URLs).

  No tool descriptions, server instructions, prompts, or resources were fetched; nothing to scan for injection.

• Supply chain

  warn25% of grade

  Package provenance: namespace verification, repo linkage, maintainer count, account age, postinstall scripts, typosquat distance.

  install hook "postinstall" present and not a recognized benign build command

• Credential hygiene

  fail15% of grade

  How the server takes secrets (env vs plaintext config vs hardcoded); secrets appearing in tool schemas.

  gateway/ai/backends/tools_infra.py: contains a hardcoded JWT (eyJh…89 (len 67)). gateway/ai/backends/tools_infra.py: assigns a literal to token (YOUR…EN (len 22)) — non-provider secret literal; prefer environment intake. Reads secrets from the environment (e.g. lib/agent.js) — the recommended intake shape.

• Permission scope

  warn15% of grade

  Declared tools vs. breadth (filesystem, network, exec); flags shell-exec and unbounded filesystem access.

  references credential paths: bin/delimit-cli.js: references a credential path (/.cursor/) references credential paths: lib/hooks-installer.js: references a credential path (/.cursor/) references credential paths: gateway/ai/integrations/opensage_wrapper.py: references a credential path (/.ssh/) has both shell/exec and network egress — capable of download-and-run, though no such path was observed (lib/agent.js: imports child_process; scripts/postinstall.js: outbound HTTP/socket call). has both shell/exec and destructive filesystem writes (lib/agent.js: imports child_process; lib/api-engine.js: destructive/writing fs.* call (write/unlink/rm/chmod/rename)). discloses shell/exec capability: 42 files (lib/agent.js, lib/api-engine.js, lib/attest-mcp.js, +39 more) discloses dynamic code execution: gateway/ai/backends/tools_infra.py: uses eval()/exec() discloses broad/destructive filesystem access: 44 files (lib/api-engine.js, lib/auth-setup.js, lib/auth-signin.js, +41 more) Capability DISCLOSURE, not a verdict: static analysis sees the primitive is present and reachable, not whether its use is attacker-controlled.

• Version behavior

  fail10% of grade

  Diff of tool definitions between versions; new permissions or changed descriptions in a patch release (the postmark-mcp class).

  vs prior 4.18.1 (patch bump): new capability primitive(s) in source: child_process, dynamic-eval, fs-write, http-client, subprocess; new hardcoded outbound host/BCC destination(s): 127.0.0.1, api.anthropic.com, api.figma.com, api.github.com, api.lemonsqueezy.com, api.npmjs.org, api.openai.com, api.telegram.org, api.x.ai, app.delimit.ai, cli.github.com, cyclonedx.org, delimit.ai, docs.delimit.ai, electricgrill.com, generativelanguage.googleapis.com, github, github.com, host, img.shields.io, jsonfeed.org, oauth2.googleapis.com, python.org, registry.npmjs.org, robotax.com, us-central1-aiplatform.googleapis.com, user, www.googleapis.com, www.npmjs.com, www.w3.org, x.com, youtu.be.

• Transport config

  fail10% of grade

  Remote servers: TLS and auth mode (none/token/OAuth). Local servers: whether the manifest indicates it phones home.

  source performs runtime self-update / download-and-execute (bypasses version_behavior); sends telemetry to third-party SDK (Datadog) — disclosure, verify opt-in; manifest/source references heartbeat behavior — disclosure