MCP server

What it is#

MCP (Model Context Protocol) is an open protocol through which agents call the tools of services. The SKILLOK MCP server makes your repositories the context of the agent you already use: it reads files and issues, searches, manages issues and proposes changes — with the same rights you have on the site.

SKILLOK is not one more chat over a model. Knowledge and code live here, and the model is a visiting worker: the subscription can be changed, but the context stays in the repositories, in git, and opens in any agent.

How to connect#

1. In your agent open «add a connector» (an MCP server) and paste the address https://skillok.org/api/v1/mcp.
2. The agent opens the SKILLOK «Allow access» page. Tick whether the agent may write and press «Allow».

That is all. The access appears in Settings → Access tokens under the agent's name, where it can be revoked. After 90 days the agent asks for permission again.

If the agent cannot sign in by itself#

Create a token in Settings → Access tokens (read only — the «read» right; issues and changes — «write» as well) and pass it in a header:

{
  "mcpServers": {
    "skillok": {
      "type": "http",
      "url": "https://skillok.org/api/v1/mcp",
      "headers": { "Authorization": "Bearer sk_…" }
    }
  }
}

Where the project context lives#

The context lives in repositories — as files, in git:

What	Where	Who sees it
Rules for the agent	the AGENTS.md file at the repository root	every agent: Codex, Cursor, Copilot, Claude Code (by importing @AGENTS.md) and SKILLOK
Project memory: decisions, notes, results	the context repository — an ordinary repository linked to the main one	the agent reads it together with the main one and writes notes there
What the agent may do	Repository settings → AI → Agents	the repository owner

The context repository is linked in the settings of the main repository. One context can serve several repositories — for example, the frontend, the backend and the documentation of a project. The agent writes notes to the context directly, and to the main repository only through a pull request.

Keep AGENTS.md short: rules, conventions, how to build and check. Reference material and the history of decisions go into the context repository: long rules get in the agent's way and cost money on every request.

What the agent may do#

The owner picks the level in Repository settings → AI → Agents:

Level	What is allowed
Read only	files, issues, search
Read and manage issues	also open issues, reply, change and close them, labels
Everything	also changes through pull requests and notes in the context repository

The level only narrows: the agent can write only if you can, and only if the token was issued with write rights. The agent cannot merge pull requests or delete anything — a person decides that.

Tools#

Tool	What it does
get_project_context	the AGENTS.md rules, what the agent may do, the context repository and its files — start here
list_repositories	your repositories and those of organizations with AI enabled
list_files, read_file, search	files, text with line numbers, search
list_issues, read_issue	issues and pull requests
create_issue, comment, update_issue	open, reply, change or close, labels
propose_change	an llm/… branch and a pull request
save_note	a note in the context repository

The server also offers the same context as the skillok://owner/name/context resource — for agents that can attach resources.

How to save money#

• Connect SKILLOK to your own agent: requests go through your subscription, not a SKILLOK key.
• Keep AGENTS.md short and the memory in the context repository: the agent reads it piece by piece when needed.
• Providers cache an identical beginning of a request (rules, tools) and charge from half to a tenth of the price for it.

Protocol versions#

2025-11-25, 2025-06-18 and 2025-03-26, without sessions: every request is a separate JSON-RPC message. Sign-in is OAuth 2.1 with PKCE (client registration or a client document at an address) or a token in a header.

Was this helpful?

Yes No