A coding agent works with Elide through two MCP servers: elide mcp inside your project, and the public docs server
at https://elide.help/mcp.
- Your editor or coding agent connects to two MCP servers.
elide mcpruns inside your project over stdio: build, run and trace tools, plus the evidence commandselide agent diagnostics,elide agent perfandelide agent bug.https://elide.help/mcpis public, over Streamable HTTP: docs search, pages, samples, comparisons, benchmarks and the report tools.- Evidence from the binary reaches a report only when the agent pastes it in. Elide never sends it.
The two servers
| Project MCP server | Docs MCP server | |
|---|---|---|
| Endpoint | elide mcp | https://elide.help/mcp |
| Transport | stdio | Streamable HTTP |
| Runs | On your machine, in the project | Hosted; public, no authentication |
| Tools | Build, run, trace, any elide command | Search, pages, samples, comparisons, benchmarks, reports |
| Details | Project MCP server | Docs MCP server |
Connect both
{
"mcpServers": {
"elide": { "command": "elide", "args": ["mcp"] },
"elide-docs": { "type": "http", "url": "https://elide.help/mcp" }
}
}Claude Code:
claude mcp add elide -- elide mcp
claude mcp add --transport http elide-docs https://elide.help/mcpEvidence from the binary
| Command | Prints |
|---|---|
elide agent diagnostics --json | Version, platform and crash-report directory |
elide agent perf --json -- <command> | Timings for a command, against another runtime or Elide version |
elide agent bug --kind <kind> --json | An elide.report/1 skeleton for bug, perf, expectation or feature |
These commands work offline and send nothing. The agent pastes their output into a report.
Report from your editor
While you work, the docs server asks your agent to notice failures, slow runs, surprises and missing features, and to offer to report them. Nothing is sent until you confirm. Accepted reports become public issues in elide-dev/elide.
- Report a problem: kinds, consent, evidence and what is stored.
- From report to fix: triage, issues and status.
- For AI agents: non-interactive flags, exit codes and commands that need a terminal.