Connect a coding agent
Opslane has a remote Model Context Protocol (MCP) server. It lets a coding agent read your latest daily summary, open an issue with its evidence, and link the pull request that fixes it.
The server accepts POST /mcp at your Opslane address. For hosted Opslane, that is https://api.opslane.com/mcp. It provides three tools:
| Tool | What it does |
|---|---|
opslane_digest | Get the latest daily summary for the project |
opslane_issue | Get one issue and its evidence from an id or URL |
opslane_link_pr | Link a GitHub pull request to an issue |
1. Create an MCP key
Section titled “1. Create an MCP key”In the dashboard, open Project settings → API keys and create an MCP key. It starts with opslane_ak_ and is shown once. The key selects one project and can read that project’s issues and link pull requests.
This key can read private data. Keep it in an environment variable, never in a committed file. The API keys guide explains expiry and revocation.
export OPSLANE_API_KEY=opslane_ak_...2. Connect your coding agent
Section titled “2. Connect your coding agent”For Claude Code, add the server from the terminal:
claude mcp add --transport http opslane https://api.opslane.com/mcp \ --header "Authorization: Bearer ${OPSLANE_API_KEY}"To share the server address with your team, commit this .mcp.json. Each person still supplies their own key through the environment:
{ "mcpServers": { "opslane": { "type": "http", "url": "https://api.opslane.com/mcp", "headers": { "Authorization": "Bearer ${OPSLANE_API_KEY}" } } }}For Codex, add the server to ~/.codex/config.toml:
[mcp_servers.opslane]url = "https://api.opslane.com/mcp"bearer_token_env_var = "OPSLANE_API_KEY"For a self-host, replace https://api.opslane.com with your own address.
Ask the agent to call opslane_digest to test the connection. A missing, invalid, expired, or revoked key returns 401. An ingest or source-map key returns 403 insufficient_scope.
Work an issue with the Opslane skill
Section titled “Work an issue with the Opslane skill”Opslane ships a Claude Code skill so the agent works a digest item the way you would. It fixes a diagnosed error on its own. For a friction issue, where the right fix is a product call, it stops and asks you which behavior you want before it writes any code.
The skill lives at examples/agent-skills/opslane/SKILL.md. Download it into your repo:
mkdir -p .claude/skills/opslanecurl -fsSL https://raw.githubusercontent.com/opslane/opslane/main/examples/agent-skills/opslane/SKILL.md \ -o .claude/skills/opslane/SKILL.mdIf you already have an Opslane checkout, copy examples/agent-skills/opslane/SKILL.md from it instead.
Then ask the agent what is broken in production, and it will read the digest, open an issue, and either fix it or ask you the product question first.
What the agent sees
Section titled “What the agent sees”Tool output is plain text. Values from a browser session, such as an error message or URL, appear between <untrusted> tags. These tags tell the agent to treat the text as data, not instructions. Opslane also limits how often each project can call the server.