Agent distribution

One product context. Every serious AI surface.

Connect Specky’s Product Graph to ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, Cursor, VS Code, and other MCP clients through one OAuth-enabled endpoint. Direct connections are available now; public directory placement follows each provider’s review process.

Canonical endpoint

https://www.specky.space/api/mcp

Streamable HTTP · OAuth 2.1 with PKCE · workspace API key fallback

Read the MCP guide

Provider paths

Where Specky can show up

A compatible MCP endpoint is the foundation. A marketplace or directory listing adds discovery, trust signals, and a native install path—but it does not replace product quality, security review, or organic content.

ChatGPT

Apps SDK app and custom MCP connector

Submit

Test Specky privately through Developer Mode, then submit an Apps SDK package to the ChatGPT App Directory.

01Use https://www.specky.space/api/mcp as the remote MCP endpoint.
02Let ChatGPT discover OAuth from /.well-known/oauth-authorization-server.
03Package three reviewable workflows and a small optional UI for the public app submission.

Prepare and submit the Apps SDK listing after the canonical deployment is live.Provider setup guide Provider requirements

Claude

Remote MCP custom connector and Connectors Directory

Submit

Connect Specky immediately as a remote MCP connector; submit the same server for directory discovery.

01Add the Specky MCP URL from Claude → Customize → Connectors → Add custom connector.
02Choose OAuth and approve a workspace-scoped consent grant.
03Submit owned endpoint, privacy, terms, examples, and maintenance details through Anthropic’s developer portal.

Run Claude smoke tests, then submit the connector directory package.Provider setup guide Provider requirements

Gemini

Gemini API remote MCP

Test

Gemini’s Interactions API can call Specky through the existing Streamable HTTP MCP endpoint.

01Configure a server-side mcp_server tool with name specky and the Specky endpoint.
02Pass a workspace bearer token through server-side credentials; never expose it in a browser.
03Restrict allowed_tools to the read-only or workflow tools required by the agent.

Add a Gemini integration example and test the read-only tool subset.Provider setup guide Provider requirements

Microsoft Copilot

Copilot Studio MCP and Microsoft 365 federated connector

Review

Copilot Studio can connect to a remote MCP server; broad Microsoft 365 availability requires connector packaging and certification.

01Add Specky under Copilot Studio → Build → Tools → Model Context Protocol.
02Use OAuth and expose a read-only tool subset for the first federated connector review.
03Prepare privacy, terms, developer identity, authentication, and certification material for Partner Center.

Create a read-only Copilot package before applying for Microsoft MCP certification.Provider setup guide Provider requirements

Perplexity

App Connectors and custom MCP

Test

Perplexity supports app connectors and MCP-based extensions; availability and public listing depend on the account surface.

01Add Specky as a trusted remote MCP connector where custom connectors are enabled.
02Use the OAuth flow so each user selects one Specky workspace.
03Lead with the feedback-to-spec and product-evidence workflows, which fit Perplexity’s research use case.

Validate a custom connector, then pursue directory or partner placement if available.Provider setup guide Provider requirements

Cursor, VS Code, Codex, and other MCP clients

Direct Streamable HTTP MCP connection

Test

The same endpoint works across compatible developer clients without a separate provider-specific backend.

01Use https://www.specky.space/api/mcp in the client’s MCP configuration.
02Use OAuth where the client supports browser authorization; otherwise use a workspace API key.
03Start with search_workspace and read-only context before enabling write actions.

Publish copy-pasteable client snippets and verify each client before changing its badge to verified.Provider setup guide

Reviewer demos

Three workflows that explain the product

Directory reviewers and new users should see a useful result in one prompt, with a clear human approval point before any consequential write.

Customer feedback to an evidence-backed PRD

“Find the strongest evidence behind this customer problem, group the signals, and draft a PRD problem statement with citations. Do not create tickets until I approve the draft.”

search_workspaceread_workspace_itemcreate_insightcreate_document

Human outcome: A cited draft lands in the PM review queue before any delivery work starts.

SaaS idea to the smallest honest experiment

“Identify the riskiest assumption in this SaaS idea, propose the smallest experiment, and create a validation page only after showing me the assumptions and stop conditions.”

start_research_sessioncreate_experimentcreate_validation_pageget_validation_page_analytics

Human outcome: The team gets a measurable validation path instead of an untested build commitment.

Approved PRD to a visible delivery handoff

“Read the approved PRD, create implementation tickets with acceptance criteria, and report the delivery handoff. Keep branch, PR, CI, deployment, and blocker state visible.”

read_approved_prdcreate_ticketshandoff_ticket_to_agentupdate_agent_run

Human outcome: The agent produces a traceable handoff while humans retain approval over execution.

Frequently asked questions

Do I need a separate Specky backend for each AI provider?+

No. Specky exposes one Streamable HTTP MCP server at https://www.specky.space/api/mcp. Providers that support MCP can use the same tool surface, while each provider may require its own directory submission or packaging process.

Will being listed in a provider directory guarantee recommendations or citations?+

No. A directory listing increases discovery and product access; it does not guarantee that an AI assistant will cite Specky, recommend it, or rank it for a query. The public guides, documentation, product evidence, and third-party authority work remain important.

Is workspace data shared across providers?+

No. OAuth grants and workspace API keys are scoped to one Specky workspace. Users should only enable the tools and provider connections they intend to use, and write actions remain reviewable in Specky.

Frequently asked questions

Do I need a separate Specky backend for each AI provider?+

Will being listed in a provider directory guarantee recommendations or citations?+

Is workspace data shared across providers?+