hc-tec/sourcerelay-agentkit/use-collector-mcp

Claude Code and agent skill by hc-tec/sourcerelay-agentkit

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Operate the Collector MCP asynchronous collection protocol safely. Use whenever an Agent must inspect Collector compatibility or browser bindings, invoke a typed Collector Tool, preserve clientRequestId idempotency, monitor an Operation, interpret exact terminal states and errors, or read bounded Artifact metadata and chunks. Pair this Foundation Skill with a platform or intent Skill for capability-specific choices.

Install

npx vskill@latest install hc-tec/sourcerelay-agentkit/use-collector-mcp

SKILL.md

Use Collector MCP

Treat Collector MCP as a thin asynchronous adapter over Collector Core. Keep planning, analysis, working notes, and final outputs in the caller; never turn MCP into a workflow service.

Establish the session

  1. Read collector://release and collector://capabilities before the first platform action.
  2. Confirm the required Tool and capability are present now. Do not infer availability from this Skill.
  3. Browser Provider Tools normally auto-select the only online session-local binding. Omit

bindingAlias in that normal case; only read collector://bindings and choose an alias when the Tool reports binding_selection_required or the caller deliberately has multiple online sessions. Official Provider capabilities intentionally do not require a binding and their Tool schemas omit bindingAlias.

  1. Do not turn provider readiness or page preparation into a separate workflow. If a capability is

absent, has no online binding, or reports runtimeState=credential_required, record that exact unavailable condition, skip that source, and continue independent sources in the caller's plan. For an Official Provider, the correct human action is to configure the provider in the local Core Gateway (or configure its documented Gateway startup environment before restarting the Gateway). MCP has no credential-configuration Tool. Never ask the user to paste a platform credential into the Agent conversation, and never manually prepare a browser tab for an Official Provider.

Never request or expose a Core token, browser binding ID, extension ID, Profile, Cookie, tab ID, URL primitive, selector, script, CDP command, or Network body.

Submit one operation

  1. Select one strongly typed Tool with the relevant Platform Skill. Confirm its Provider and binding

requirement from the live capability contract, but do not make a binding/page-preparation call a prerequisite when the Tool can resolve it internally.

  1. Generate one UUID clientRequestId for this exact canonical Tool call. Omit bindingAlias for

the ordinary single-online-binding case; include it only when an explicit session choice is needed. On Windows, carry non-ASCII arguments across an explicitly UTF-8-safe boundary (for example a Unicode environment variable or decoded UTF-8 bytes), never a Windows PowerShell 5.1 native text pipe. Confirm the intended UTF-8 bytes before the Tool call; a query changed to ? is a different, invalid platform action and must not be submitted.

  1. Preserve the complete Tool name and arguments until the outcome is known.
  2. Call the Tool once. Do not poll inside the Tool call or issue a second capability implicitly.
  3. Record the returned operationId, operationResourceUri, capabilityId, and

idempotentReplay in caller-owned state.

For a genuinely new platform action, generate a new request ID. For the same action, never change arguments while reusing an ID.

Follow the operation

Read the returned Operation Resource until it reaches a terminal Core state or the caller's explicit deadline. Use bounded local polling such as 2, 4, 8, then at most 15 seconds between reads; do not busy-loop.

  • Treat queued and claimed as nonterminal facts.
  • Treat completed, partial, stopped, and failed as exact terminal facts.
  • Preserve coreState, terminalReason, and errorCode; never rewrite failure as no_results.
  • Use an Artifact attached to partial, but retain the partial status and coverage gaps.
  • Stop on login, captcha, risk, rate-limit, permission, or safety terminals. Do not bypass them.

For submission_outcome_unknown, make no automatic retry. If reconciliation is necessary, explicitly resubmit the exact same Tool arguments once with the same clientRequestId; then stop if uncertainty remains. For submission_conflict, do not silently replace the ID.

Read an artifact

  1. Read collector://artifacts/{artifactId} before content.
  2. Verify availability, media type, byte length, whole SHA-256, capability, and terminal status.
  3. Start at collector://artifacts/{artifactId}/chunks/0 only when content is needed.
  4. Follow only the returned nextChunkResourceUri; stop at null or once enough evidence is read.
  5. Preserve Artifact ID, hash, byte range, and truncation with any caller-owned quotation or summary.

Do not invent file paths, cursors, JSONPath, byte limits, or alternate Artifact identities. Do not ask MCP to summarize content or call a model.

Apply the contract reference

Read references/protocol.md when interpreting an error, designing polling, or recording Resource provenance. Keep capability-specific decisions in the Bilibili or Xiaohongshu Platform Skill.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/hc-tec/sourcerelay-agentkit/use-collector-mcp/badge)](https://verified-skill.com/skills/hc-tec/sourcerelay-agentkit/use-collector-mcp)

Something wrong with this skill or its page? Tell us.