Loading ToolBench

Reading the current corpus…

prompt-capsule MCP server · C · ToolBench

prompt-capsule

MCP server for sealing prompts/text into one-time tape URLs with TTL, supporting create, read, and delete operations. Includes program codes (stable aliases), moderation, and telemetry.

2026-07-28+STREAMABLE_HTTP3 tools

63/100

Definition Quality74

Protocol Readiness75

Supportability35

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · evidence: stateless requests, Streamable HTTP

Current-spec patterns detected

Summary

Three tools with clear naming (verb_noun pattern) and documented schemas. Descriptions are present but vary in quality, create_prompt_tape is detailed (180+ chars) with constraints clearly stated; read_prompt_tape is terse (70 chars); delete_prompt_tape lacks context on consequences. All tools have input schemas with types and descriptions. Output schemas are not explicitly documented in the source. Error handling is minimal, no recovery guidance or actionable error messages visible. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles (WRITE, READ_ONLY, DESTRUCTIVE).

Tools (3)

create_prompt_tapewritesource verified85/100

Seal a prompt/text into a one-time tape URL. content ≤16KB; ttl_seconds ≤604800 (7 days, default 7 days). Returns view_url (for humans), raw_url (for an agent to fetch), code (the tape code = slug; downstream can pass just the code to read_prompt_tape), code_share_text (a URL-free share line carrying only the code, for platforms that downrank links), delete_token, expires_at, and agent_text.

delete_prompt_tapedestructivesource verified67/100

Delete a tape using its delete_token.

read_prompt_taperead onlysource verified70/100

Read a tape body. target can be a tape code (slug), a program code (a long-lived, case-insensitive alias), or a full URL (/c/… or /view/…).

Issues (5)

HIGHtool ↗

Output schemas not documented. LLMs cannot plan downstream calls or extract required fields (e.g., code, view_url, delete_token from create_prompt_tape). Responses must be structured and typed.

create_prompt_taperead_prompt_tapedelete_prompt_tape

HIGHtool-description ↗

read_prompt_tape description is 70 chars, below the 100-char minimum for clarity. Does not explain when to use it vs. other tools, what it returns, or how target resolution works (code vs. URL vs. program code).

read_prompt_tape

HIGHcommand-tool ↗

delete_prompt_tape description (40 chars) is critically short and lacks context on consequences. No mention that deletion is irreversible or that the delete_token is required and sensitive.

delete_prompt_tape

MEDIUMtool ↗

No tool annotations despite clear risk profiles. create_prompt_tape is WRITE, read_prompt_tape is READ_ONLY, delete_prompt_tape is DESTRUCTIVE. Annotations (readOnlyHint, destructiveHint) help LLMs reason about safety and idempotency.

Recommendations

Document output schemas for all three tools. For create_prompt_tape, explicitly list: view_url (string), raw_url (string), code (string), code_share_text (string), delete_token (string), expires_at (ISO 8601 string), agent_text (string). For read_prompt_tape, document the response body structure. For delete_prompt_tape, document success/failure response format.
Expand read_prompt_tape description to 100 - 150 chars. Explain: 'Retrieve the sealed prompt/text body from a tape. Target can be a tape code (slug), program code (long-lived alias), or full URL (/c/… or /view/…). Returns the original content and metadata.'
Rewrite delete_prompt_tape description to emphasize irreversibility and security: 'Permanently delete a tape using its delete_token. This action cannot be undone. The delete_token is a one-time secret, do not share it. Returns success or error.'
Add tool annotations: create_prompt_tape with destructiveHint=false (it creates, not destroys), read_prompt_tape with readOnlyHint=true, delete_prompt_tape with destructiveHint=true and idempotentHint=false (repeated deletes with same token should fail, not silently succeed).
Add error handling guidance to descriptions. E.g., create_prompt_tape: 'Returns 400 if content exceeds 16KB or ttl_seconds exceeds 604800. Returns 429 if rate-limited (10 creates/min per IP). Returns 500 on server error, retry with exponential backoff.' Similar for read and delete.

Score history

Overall score trend

First recorded score · v2 rubric

63/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	63	2026-07-28+	v2