Connect VS Code (GitHub Copilot agent mode) to Willder
Copilot's agent mode in Visual Studio Code.
Transport: Streamable HTTP · Auth: Authorization: Bearer <capability token> ·
Endpoint: https://willder.ai/api/mcp
Half of this row is verified, and the two halves are different claims. The server's half passes: on 2026-10-05, against deployed commit 2eed3f05, npx tsx scripts/mcp-client-probe.ts --client=vscode-copilot --transport-only reached https://willder.ai/api/mcp using Streamable HTTP with the exact Authorization header this page commits, and the initialisation sequence, tools/list and one scoped memory_query all answered, with an absent token and a wrong token both refused. The client's half is unrun. Nobody has run this client against Willder, so nothing here claims it works. Why: the Copilot MCP panel is only visible inside VS Code. That needs a person's run recorded in manual-runs.json, naming who ran it, when, and on which commit — see the index.
Get a token
Open Access → Tokens in your Willder workspace and issue a capability token for the narrowest scope that answers your questions. The value is shown once; only its hash is stored. Keep it in the client's secret storage or an environment variable — never in a committed file, a prompt or a screenshot.
export WILLDER_TOKEN='<your capability token>'Config file
.vscode/mcp.json in the workspace, or the user mcp.json via MCP: Open User Configuration.
{
"inputs": [
{ "type": "promptString", "id": "willder-token", "password": true, "description": "Willder capability token" }
],
"servers": {
"willder": {
"type": "http",
"url": "https://willder.ai/api/mcp",
"headers": { "Authorization": "Bearer ${input:willder-token}" }
}
}
}${input:...} prompts once and stores the value in VS Code's secret storage, so the committed
file holds no token. ${env:WILLDER_TOKEN} also works if you prefer the environment.
Name your task (optional). Add "X-Willder-Task": "${env:WILLDER_TASK}" to headers and
export WILLDER_TASK (a ticket key or branch name) before starting VS Code: the first result of
each session then leads with what earlier sessions of that task decided and wrote.
Verify the connection
Run MCP: List Servers and open willder; it should report running with its tools. Then use a Willder tool from agent mode.
Notes
VS Code falls back to SSE only if the server asks for it; "type": "http" keeps it on Streamable HTTP.
See the index for the other clients and mcp/README.md for the tool surface and the token rules.