To connect 1PC to Kimi Code for VS Code, add https://use1pc.com/mcp as a remote HTTP MCP server, complete 1PC authorization and test the connection. Then explicitly enter Product Owner, read the selected product and verify one authorized save in a new conversation.
This guide uses the VS Code extension's documented MCP interface, reviewed on 4 October 2026. It is a setup procedure to check in your installed version, not a claim that a live Kimi-to-1PC run has already passed. For the broader product workflow, read Kimi for Founders.
Prepare the host and your product workspace
Install Kimi Code for VS Code, open a workspace folder and configure the extension's own account or API access. Those prerequisites are described in the official Kimi Code quick start. A Kimi API key authenticates the host; it is not a 1PC MCP credential.
Create or sign in to your 1PC account and complete your founder profile. You can reuse an existing product after selecting it in the connected session. Tool discovery and founder-context reads do not require you to create a duplicate product.
For a first connection, choose a small task on a product you may access. The 1PC MCP reference documents current tools and permissions. The MCP overview explains how a compatible client and saved product state fit together.
Add a remote HTTP server and authorize 1PC
The official customization guide places MCP Servers in the extension's gear menu. It documents remote http, optional OAuth, an authorization action and a connection test.
- Open the gear menu and choose MCP Servers.
- Add a remote http server with URL
https://use1pc.com/mcp. 1PC uses Streamable HTTP. - Use the OAuth authorization action. In the browser, sign in to 1PC, inspect the requested access and grant the scope needed for your task.
- Return to the extension, test the connection and save the server configuration when the test succeeds.
The menu and transport instructions follow Kimi's documentation; the endpoint and access rules are 1PC's. If the authorization flow cannot register the client or finish its callback, check 1PC's authentication section rather than guessing extra configuration fields.
1PC uses knowledge:read for reads and discovery. Saving also requires knowledge:write. Website login cookies do not authenticate MCP. A 1PC personal token is another option for clients that support bearer headers, but this guide's VS Code path uses OAuth. Keep credentials in the host's supported credential storage.
Make Product Owner the explicit entry point
A working connection exposes 1PC tools. It does not automatically register the plugin's native Product Owner skill in Kimi. In a host without that skill, use a bootstrap request that loads the backend entry requirements and selects the actual product before domain work:
Use the connected 1PC tools. First load get_skill_instructions
with skill_id "product-owner" and phase "entry".
Read get_workspace_context and paginate list_product_skills.
List my products and let me select the relevant one; do not create
a duplicate. Load get_product_context with that product_id.
Explain its saved state, current access and unfinished work.
This first check is read-only. Do not save or start another workflow.
Use the current tool schemas the server exposes, including the selected product ID for product calls. This request is public setup guidance, not the body of a built-in execution skill. Keep the entry protocol available for later 1PC requests using whatever instructions your host supports; a new session should load the current entry guidance again.
For a requested task, Product Owner matches the intent to the paginated catalog. The assistant loads the selected skill through get_skill_instructions with phase work, then follows its separate requirements. Backend Markdown supplies task guidance; it is not a native Kimi skill installation.
Prove a saved continuation with a small task
Connection status alone does not prove that the assistant can complete a product task. Use a product you control and authorize a small factual update. Keep the full check visible:
| Check | What establishes success |
|---|---|
| Discover and read | The client discovers tools and reads your founder context and selected product's actual saved state. |
| Save an authorized result | The assistant follows the task requirements, preserves existing information and reports a successful save receipt. |
| Read it back | The assistant reads the changed file and verifies the content and revision. |
| Continue in a new chat | A fresh conversation selects the same product and retrieves that saved result without relying on the old transcript. |
| Check access changes | Read-only consent prevents saving; after revocation, authenticated workspace reads fail. |
Follow the returned destination and revision checks for the update. If you only asked for status, do not turn the check into a saved plan. Resolve explicit access or revision errors before retrying. If a write response is uncertain, read the destination first: the update may already have been saved.
Record the Kimi extension version and date with your result. The cross-client continuity guide explains why a fresh read is stronger evidence than an assistant saying it remembers the task.
Locate the failure before changing the workflow
- No usable remote connection: verify the URL and HTTP transport, then check the extension's connection result. A local stdio configuration is a different path.
- Authorization does not finish: inspect consent and callback handling. Consult the 1PC authentication reference for registration requirements.
- HTTP 401: authorization may be missing, expired or revoked. Reauthorize the intended account.
- Reads work but writes fail: inspect OAuth scope, product membership and current plan. A Kimi subscription does not change 1PC entitlements.
- The assistant cannot find a skill: discover the catalog and paginate it. Private skills must be published in the caller's workspace; a draft is not available for execution.
- A new chat lacks the result: confirm the same account, product and saved file. Conversation-only work is not carried into 1PC automatically.
The MCP error and recovery reference covers server-side responses. A host with tool limitations may need a smaller task or a different compatible client; inventing a successful tool call does not resolve the limitation.
Use the connection for the next product decision
With read access verified, start a weekly product review. With the appropriate paid access, collect an existing product's context, draft a feature-launch marketing plan or apply a published private skill.
Choose one bounded task and verify its saved outcome. Publication, prospect contact and deployment need their own authorization and available host tools.
Keep the authentication and tool reference nearby.
Use the 1PC quickstart to check current scopes, tool availability and the read/save/read-back contract while you connect.
Open the 1PC MCP quickstart ↗