flowstate-mcp-web-transport Package Skill

Package: @epicdm/flowstate-mcp-web-transport Repository: epic-flowstate-community Path: packages/mcp-web-transport Version: 1.0.7

Use this published Dojo skill as the package-level entrypoint. For detailed workflows, use the package-local skills listed below.

Primary Package Workflow

MCP Web Transport Chat And Invoke

Status: Active Purpose: Use MCPWebTransport correctly for chat and Phase 1 text-wrapped tool invocation. Scope: src/transport/MCPWebTransport.ts. Trigger: A task calls or changes MCPWebTransport.chat() or MCPWebTransport.invoke(). Input: Messages or tool name/arguments, provider config, optional system prompt, optional tools, and optional chunk callback. Output: Correctly handled MCPResult.

Choose The Method

Chat Example

import { MCPWebTransport } from '@epicdm/flowstate-mcp-web-transport'

const transport = new MCPWebTransport()
const result = await transport.chat({
  messages: [{ role: 'user', content: 'Summarize this task.' }],
  agentId: 'agent_123',
  conversationId: 'conv_123',
  provider,
  systemPrompt: 'Be concise.',
})

if (!result.success) {
  throw new Error(result.error)
}

Invoke Example

const result = await transport.invoke({
  toolName: 'task-list',
  arguments: { assignee: 'me' },
  userId: 'user_123',
  agentId: 'agent_123',
  conversationId: 'conv_123',
  provider,
})

invoke() sends a user message containing Execute tool: ... and serialized arguments. Use MCPToolExecutor for actual gateway tool calls.

Error Handling

Provider errors are caught and returned:

if (!result.success) {
  log.warn({ err: result.error, latencyMs: result.latencyMs })
}

Verification

yarn workspace @epicdm/flowstate-mcp-web-transport test -- MCPWebTransport
yarn workspace @epicdm/flowstate-mcp-web-transport typecheck

Red Flags

Created: 2026-06-29

Package Skill Inventory

@epicdm/flowstate-mcp-web-transport Skill Inventory

Inventory date: 2026-06-29 Package path: packages/mcp-web-transport Dojo status: local skills authored; publication deferred until package skill review and repo-wide Dojo sync.

Skill Map

Composition

Start with flowstate-mcp-web-transport. Use focused skills together for a real tool-enabled chat flow:

1. flowstate-mcp-web-transport-provider-factory
2. flowstate-mcp-web-transport-mcp-tool-executor
3. flowstate-mcp-web-transport-streaming-tools
4. flowstate-mcp-web-transport-chat-invoke
5. flowstate-mcp-web-transport-chat-service
6. flowstate-mcp-web-transport-maintenance

Review Notes

The generated Dojo skillset under .flowstate/dojo/agent-skillset.md is not sufficient for publication as-is. It names feature exports but does not teach provider selection, text-wrapped invoke limitations, streaming chunk semantics, MCP Gateway dependency behavior, or maintenance drift.

Review Gate

@epicdm/flowstate-mcp-web-transport Skills Review

Review date: 2026-06-29 Reviewer: Codex Status: reviewed with verification notes

Evidence Used

• packages/mcp-web-transport/package.json
• packages/mcp-web-transport/README.md
• packages/mcp-web-transport/.flowstate/docs/index.md
• packages/mcp-web-transport/.flowstate/docs/api/index.md
• packages/mcp-web-transport/.flowstate/docs/examples/basic-usage.md
• packages/mcp-web-transport/src/index.ts
• packages/mcp-web-transport/src/types/index.ts
• packages/mcp-web-transport/src/providers/factory.ts
• packages/mcp-web-transport/src/providers/openai.provider.ts
• packages/mcp-web-transport/src/providers/anthropic.provider.ts
• packages/mcp-web-transport/src/transport/MCPWebTransport.ts
• packages/mcp-web-transport/src/services/MCPToolExecutor.ts
• packages/mcp-web-transport/src/services/ChatServiceWithTools.ts
• packages/mcp-web-transport/src/**/__tests__/*.test.ts
• packages/mcp-web-transport/.flowstate/feature-matrix/inventory.md
• packages/mcp-web-transport/.flowstate/dojo/agent-skillset.md

Quality Gate

• Source-backed purpose: pass
• Public entry points captured: pass
• Package connections captured: pass
• Composable skill inventory: pass
• Skills formatted with name and trigger-focused description: pass
• Generated-skill replacement quality: pass
• Verification commands run: pass with warnings noted

Drift Findings

• README.md usage contains a placeholder import instead of practical examples.
• The package description/README promise HTTP/SSE transport, but source implements provider SDK calls with callback streaming.
• MCPToolExecutor imports @epicdm/flowstate-mcp-client, but package.json does not declare it.
• package.json declares peer dependency @epicdm/flowstate-agents-core, but source does not import it.
• Tests are mostly interface/shallow provider tests and do not fully mock provider streaming or tool-call execution.
• ChatServiceWithTools builds tool-result follow-up messages but does not make a second provider call before returning.
• Generated Dojo skillset is export-list oriented and not ready to publish.

Verification Results

• ASCII scan for package-local skills: passed.
• Skill frontmatter scan for all SKILL.md files: passed.
• yarn workspace @epicdm/flowstate-mcp-web-transport test --runInBand --no-cache: passed, 5 suites, 13 tests.
• yarn workspace @epicdm/flowstate-mcp-web-transport typecheck: passed.
• yarn workspace @epicdm/flowstate-mcp-web-transport build: passed with existing package export condition warning for types after require/import.
• yarn workspace @epicdm/flowstate-mcp-web-transport lint: passed with 112 existing warnings.
• yarn nx build @epicdm/flowstate-mcp-web-transport: passed, building flowstate-mcp-client and agents-core dependencies; emitted existing export-order warnings and Nx Cloud unconnected-workspace 401 notice.

Publication Decision

Do not publish the existing generated Dojo skilllets. Publish only after local package skill verification and the repo-wide Dojo sync step convert these reviewed skills into Dojo payloads.

Start with this skill to understand the package purpose, then use the package-local focused skills for specific workflows. Keep this Dojo version aligned with packages/mcp-web-transport/package.json. Do not treat generated feature lists as lessons; use workflow evidence from CODE-REVIEW.md, SKILL-INVENTORY.md, and REVIEW.md.