Step 2: Wire GitHub and Linear Connections

What it is#

A connection wires the agent into an external MCP server — here GitHub and Linear. Eve discovers the remote tools and surfaces them to the model as qualified names (github__*, linear__*). Tokens resolve out of band and never reach the model. Connections live under agent/connections/; the runtime name is the filename.

Why you'd care#

The digest needs real activity: recently merged PRs, opened issues, cycle progress. Hand-rolling a GitHub and Linear client means auth, token refresh, schemas, and error mapping — twice. A connection collapses each into one file. The model gains github__list_pull_requests, linear__list_issues, and friends with zero integration code in your tools.

Before#

The agent only has its persona and any tools you authored. To read GitHub or Linear you'd wrap their APIs by hand.

After#

Linear — app-scoped token via getToken:

import { defineMcpClientConnection } from "eve/connections";

export default defineMcpClientConnection({
  url: "https://mcp.linear.app/mcp",
  description: "Linear workspace: issues, projects, cycles, and comments.",
  auth: {
    getToken: async () => ({ token: process.env.LINEAR_API_TOKEN! }),
  },
});

GitHub — same shape, its MCP URL:

import { defineMcpClientConnection } from "eve/connections";

export default defineMcpClientConnection({
  url: "https://api.githubcopilot.com/mcp",
  description: "GitHub: repositories, pull requests, issues, and commits.",
  auth: {
    getToken: async () => ({ token: process.env.GITHUB_TOKEN! }),
  },
});

The model now sees linear__list_issues, github__list_pull_requests, github__list_commits, etc. Tool names are the connection prefix plus the remote tool name, joined by __.

Do it yourself#

1. Create agent/connections/linear.ts and agent/connections/github.ts with the examples above.
2. Add LINEAR_API_TOKEN and GITHUB_TOKEN to .env (never commit secrets).
3. Run eve dev and ask "List my open Linear issues and recent GitHub PRs." Confirm the model calls linear__list_issues and github__list_pull_requests.
4. Try "Summarize what changed in the last 24 hours across both." The model should connection_search then call the qualified tools.
5. Leave the research depth shallow for now — the researcher subagent takes over deep-dives in Step 3.

Gotchas#

• App-scoped auth for background work. Schedules and local dev have no end-user principal, so connect("linear/myagent") (user-scoped) fails principal_required there. Use principalType: "app" (the getToken form above) for agent-owned reads.
• Tokens never reach the model. The URL and credential are brokered by eve; they stay out of conversation history.
• Qualified names are the contract. The model calls github__list_pull_requests, not list_pull_requests. If it drops the prefix, the call fails. Point it at the connection name in your instructions.
• Per-connection approval optional. Read-only digests rarely need it. Add approval: once() from eve/tools/approval only if you later let the agent write.

Recap#

Two defineMcpClientConnection files give the root read access to GitHub and Linear by qualified tool name, with tokens brokered out of band. Next: add the researcher subagent that deep-dives on each item.