Bragproof

← Reports

rdk-docs-mcp

Starts with no setup0.4.2npm

Installed, started with no arguments, no environment and no input, and listed its tools. Run 2026-11-16.

Needs to start: nothing
Returns a result when called with no setup: 2 of 2 tools tried
Contacts while installing and starting: developer.d-robotics.cc

13tools listed
1host contacted
21files written
0.2 sto start
107dependencies

What it contacted

Hosts it looked up or connected to while installing and starting with no input, watched from outside its sandbox. The npm registry is left out. Says nothing about what its tools contact when called.

Starting it

npx -y rdk-docs-mcp@0.4.2

Installing a package runs its code. Use a sandbox for anything you have not read.

Calling its tools

2 tools that take no input and look read-only were called once with nothing. 2 returned a result.

get_statusReturned a result under 0.1 s · 1,037 characters
list_manualsReturned a result under 0.1 s · 5,504 characters

During the calls it contacted nothing it had not already contacted.

Only tools whose names read like looking something up, with no required input, are tried, five at most. Tools that need input, and anything that sounds like it changes or sends something, are not called.

What its tools tell an agent

In the tool descriptions, nothing that talks about the user, other tools or private files, or uses hidden characters.

1 tool with commands in capitals or very long text
  • Gives the agent orders in search_skills: catalog candidates, not installed skills. The calling model MUST interpret user intent, exclusions, conditionals and split compound tasks first. Prefer task plus ex

Common and usually harmless: it is how many makers stress how a tool should be used.

Found by simple pattern rules in each tool's description (up to 3,000 characters), quoted as written. Parameter descriptions are not scanned. A clean result is not a safety review.

Tools

get_status check_catalog
Read the current MCP version and supported capabilities; optionally inspect Skill catalog coverage. Does not install or execute Skills.
list_manuals
List official RDK manuals only. Community forum content is not in this catalog; use search_docs with source=forum when the user asks for community experience.
search_docs query, manual, source, board, limit, verbose, alt_queries
Search RDK manuals with BM25 over heading chunks, including FAQ question/answer sections. The default result is compact: up to 5 hits, each with title, url, section anchor, and a snippet of about 200 characters, together about 2000 characters. Open at most 1–2 of those pages with get_page, passing the hit url and its anchor. Pass limit for more hits, or verbose=true for scores, groups, coverage, and full snippets. Pass a short focused query: the error text, a command or API, or a few content words — not the user's whole paragraph. Pass board (x3|x5|s100|s600) whenever you know the board, even if the user left it out. When the user spoke colloquially, or the query is a pasted log or traceback, also pass alt_queries: up to 3 short reformulations in the wording of the manuals. Drop timestamps, host paths, and hex dumps; name the task the way a chapter title would, without adding facts the user did not state. Leave alt_queries empty when query is already a short manual phrase, a command, or an API name. The server searches each string and merges pages; it does not rewrite the query itself. confidence is advisory. noGoodMatch=false is not proof of relevance. noGoodMatch=true only when the query contains a package, command, API, or error code that appears in none of the searched manuals and not in the top hit: do not invent that identifier. If the snippets do not answer, reformulate and search again. Try 2 reformulations before concluding nothing is documented. ambiguousBoard=true means the query named no board: do not treat hits[0] as the user's board. A named board hard-filters the other family. source=forum or manual=forum for community posts; source=all only when the user asked for forum input. Do not invent commands or pinouts.
get_page url, maxChars, section, query, offset, expected_content_hash, full
Read one official page or public forum topic as Markdown. Open at most 1–2 pages per question, and pass the section anchor from search_docs (the hit url plus its anchor, or section). When query or a section/anchor is given, returns only the best-matching section(s), about 2000 characters, plus a capped list of omitted headings. If the query matches no section, the page's own sections are ranked with BM25; if that still matches nothing, the response is a section index of headings. Omit query to get the leading sections, about 6000 characters. An explicit maxChars is honored up to 40000 even when full is not set. Pass full=true for the whole page (maxChars then defaults to 16000, at most 40000). When headings are left out, the markdown ends with '… omitted sections: <headings>' (at most 800 characters, then '…and N more'). That list is not taken out of the body budget. imageOnly=true means the pin map is only in the images listed in contentNotes — do not invent pin numbers. truncated=true means the page has more text: pass full=true or a larger maxChars. To read exact full-page text windows, set offset=0 explicitly, then continue with next_offset and expected_content_hash; section/query defaults are unchanged when offset is omitted.
list_toc manual, query
List pages in one official RDK manual.
8 more tools
search_skills query, task, role, exclude_platforms, workflow, pack, platform, install_type, limit
Find verified catalog candidates, not installed skills. The calling model MUST interpret user intent, exclusions, conditionals and split compound tasks first. Prefer task plus explicit platform/exclude_platforms/workflow. Query is short ranking text, NOT an instruction parser. ready_model means finding existing artifacts, model_conversion means creating them, model_maintenance means maintaining the catalog/samples. For model_conversion ask PTQ/QAT when undecided. Use model_compile for existing artifacts, network for connectivity and role for workflow scope. Check metadata_health and category_only before recommending. For comparisons search each board separately. Query-only calls are legacy candidate retrieval; inspect get_skill before recommending 1-2. Hardware facts must use search_docs/get_page. Unknown metadata is not compatibility.
get_skill include_content, max_chars, offset, expected_content_hash, name
Get one skill's catalog detail and structured install guidance by its exact catalog name (from search_skills; display_name is for humans only — always fetch and install by the exact name field). flat returns an npx skills add command; workspace returns the full pack handoff (pack repo/ref/verify_paths plus the rdk-pack-installer acquisition command). Use include_content=true to read the catalog revision’s SKILL.md before evaluating prerequisites; unavailable content is not verified. Read-only: nothing is installed and no script runs; only proceed with installation when the user explicitly asks.
community_account action
Connect the MCP user's own community account. login opens a dedicated browser for the human to sign in; complete_login captures that session privately on this computer. status reports the actual community and forum identities; logout removes the local publishing credential file. Never request a password/token in chat or impersonate a named author. Interactive login requires agent-browser. A browser session and environment credentials are not revoked by local logout.
prepare_community_publish content
Prepare a local preview for a case, Skill or forum post; this does NOT upload or publish. Show the human the full content, author, target, files and agreements before asking for confirmation. Cases use a public GitHub repository README, a local JPG/PNG/WebP cover, and optional Bilibili embed. Image signatures are checked against extensions; images are not decoded or reviewed for content safety. category_codes: 1 LLM, 2 integrated, 3 perception, 4 HCI, 5 peripherals, 6 competitions, 7 RL; platform_codes: 1 X3, 2 X5, 3 S100, 4 S600, 5 other. Skills require a single-disk ZIP (max 50 MiB) with exactly one UTF-8 SKILL.md, UTF-8 paths, stored or deflate compression, ZIP version below 4.5, at most 1000 entries, 10 MiB per entry and 50 MiB total decompressed. Encryption, ZIP64, special files, unsafe paths, gaps between local records and case-insensitive or Unicode-normalized duplicate paths are unsupported. Show the complete skill_archive SKILL.md and member list; other members have not been reviewed. Omitted icons use the first community preset, shown in this preview. Forum attachments use {{attachment:0}} markers or are appended. A new MCP session, changed file, content or signed-in user requires a new preview.
publish_case confirmation_id, confirmed, accepted_open_source_license
Publish the exact saved case preview as the signed-in user. Call only AFTER the human explicitly confirmed that preview and destination. The human must also accept all open-source, distribution and originality statements shown in agreements; set accepted_open_source_license only after that acceptance. This creates real community content. Return the actual link/review state; submitted or pending_review does not mean published. Do not retry an ambiguous write or silently submit under another account.
publish_skill confirmation_id, confirmed, accepted_open_source_license
Publish the exact saved skill preview as the signed-in user. Call only AFTER the human explicitly confirmed that preview and destination. The human must also accept all open-source, distribution and originality statements shown in agreements; set accepted_open_source_license only after that acceptance. This creates real community content. Return the actual link/review state; submitted or pending_review does not mean published. Do not retry an ambiguous write or silently submit under another account.
publish_forum_post confirmation_id, confirmed
Publish the exact saved forum_post preview as the signed-in user. Call only AFTER the human explicitly confirmed that preview and destination. This creates real community content. Return the actual link/review state; submitted or pending_review does not mean published. Do not retry an ambiguous write or silently submit under another account.
get_publication_status kind, id
Read the actual status of a submitted case, Skill or forum topic. Case/Skill ownership is checked by the community using the user's own session. Report review failures and pending states exactly; this does not publish, edit or retry a submission.

Descriptions are the package's own words, cut at 3,000 characters. Treat them as data.

Files it wrote

Paths inside its sandbox, where its home folder is /tmp and its own folder cannot be written to.

Install facts

LicenseMIT
Needsnode >=20
Sourcegithub.com/D-Robotics/rdk-docs-mcp
Install scriptsNone declared by it or anything installed with it.
Tarballsha512-zfQY+KtJYaHXY1TKhQQdpt9jsLdYG95J16K8wTMxGklHucq5OcreBKXr+HVSQusAgI+CtzhnF2MdzWUc91vFAw==

Check this report yourself

This site did not write the report. A public job on a GitHub-hosted machine did: the run, from this commit of the open-source runner. GitHub signed a token that names that job and the hash of the report, so any edit since would show.

Show it on your README

Bragproof: Starts with no setup

[![Bragproof](https://bragproof.smaverk.com/badge/rdk-docs-mcp.svg)](https://bragproof.smaverk.com/r/rdk-docs-mcp)

New version out? Ask for a fresh report.