Desktop, Skills & MCP

Give Kendr the right instructions and tools.

Skills teach Kendr how to handle a repeatable job. MCP servers give Kendr access to trusted tools and data. Install a skill pack from Kendr Desktop, connect remote MCP servers from your account, and decide exactly which capabilities can be used.

33 available packs Desktop skills Remote MCP tools Approval-aware actions

Skills and MCP are different

SkillMCP server
PurposeGives Kendr workflow instructions, domain guidance, checks, and expected output.Exposes callable tools and external data through the Model Context Protocol.
ExampleA postmortem skill tells Kendr how to structure impact, timeline, causes, and actions.A Grafana MCP server can query dashboards, logs, alerts, or incidents.
Where to add itInstall a skill pack from Skills → Marketplace in Kendr Desktop.Add a remote endpoint under MCP servers, or install a desktop pack that bundles a local server.
Security boundaryReview what the skill instructs Kendr to do.Trust the server, limit allowed tools, protect its token, and review consequential actions.
Use both together

A pack can combine workflow instructions with one or more MCP servers. The skill explains when and how to work; MCP supplies the tools.

Install a skill pack in Kendr Desktop

  1. 1
    Open Kendr Desktop
    Sign in with the Kendr account you use for Cloud and API access.
  2. 2
    Open Skills → Marketplace
    Search by pack name, workflow, tool, or category.
  3. 3
    Review the pack
    Read its description, included skills, MCP servers, tool count, version, source, and any credential requirements.
  4. 4
    Install and enable
    Install the pack, then enable the pack or only the individual skills you want Kendr to use.
  5. 5
    Provide required connections
    If the pack needs a service token, local runtime, or app connection, complete that setup before the first task.
  6. 6
    Run a matching task
    Describe the outcome in Chat or agentic mode. Kendr can select the enabled skill when its instructions match your request.
Updates

Kendr Desktop compares installed pack versions with the hosted catalog and can offer an update when a newer version is available.

Connect a remote MCP server

Use this flow for a trusted MCP endpoint that supports Streamable HTTP or HTTP/SSE over HTTPS.

  1. 1
    Get the official endpoint
    Copy the public MCP URL from the server provider. Confirm the hostname and documentation before entering a token.
  2. 2
    Open MCP servers
    Sign in and open Account → MCP servers.
  3. 3
    Describe the server
    Enter a human-readable name, a unique server label, its HTTPS endpoint, and a description of when Kendr should use it.
  4. 4
    Limit tools when possible
    Enter a comma-separated allowlist such as search, fetch. Leave it empty only when every tool exposed by the server is appropriate.
  5. 5
    Add authorization
    If required, paste the OAuth access token or bearer token. Kendr stores it separately and does not display the full value again.
  6. 6
    Trust and enable
    Confirm that you trust the server and its tools, then enable it for Chat.
  7. 7
    Activate it in Chat
    Open Chat → Tools, select the server, and run a read-only request first to confirm the connection.
Local servers

The Cloud connection screen requires a public HTTPS endpoint. Local stdio servers are normally supplied through Kendr Desktop skill packs and run on the computer where Desktop is installed.

Use it from the API

A registered custom server is callable through the hosted model API with {"type": "kendr_mcp", "server_id": "mcp_..."}. This is intentionally separate from OAuth app connectors, which use {"type": "kendr_app", "selection": "auto"} for authenticated discovery and relevance selection. MCP servers must be enabled and trusted; Kendr injects stored authorization server-side. See Call remote MCP tools.

MCP connection fields

FieldWhat to enter
NameA clear label for people, such as “Company knowledge” or “Production Grafana.”
Server labelA stable identifier using letters, numbers, underscores, or hyphens, such as company_knowledge.
MCP endpointThe complete provider URL beginning with https://.
DescriptionWhat the server provides, when Kendr should use it, and important boundaries.
Allowed toolsAn optional allowlist of exact tool names. Use it to keep unused or high-risk tools unavailable.
Bearer tokenAn optional service token. Enter it only on Kendr's secure connection screen.
TrustedYour confirmation that you recognize the provider and accept that its tools may receive relevant conversation context.
EnabledWhether the server can currently be selected in Chat.

Connect MCP through the API

Signed-in applications can create the same account-scoped MCP configuration with POST /api/me/mcp-servers.

curl https://kendr.org/api/me/mcp-servers \
  -X POST \
  -H "X-Kendr-Session: $KENDR_SESSION" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Company knowledge",
    "server_label": "company_knowledge",
    "server_url": "https://mcp.example.com/mcp",
    "description": "Search approved company documentation.",
    "allowed_tools": ["search", "fetch"],
    "authorization": "YOUR_BEARER_TOKEN",
    "trusted": true,
    "enabled": true
  }'

List configurations with GET /api/me/mcp-servers. Update or delete one server at /api/me/mcp-servers/{server_id}. Treat the authorization value as a secret and never log the request body.

Use MCP tools safely

  • Connect only servers whose publisher, hostname, permissions, and privacy policy you trust.
  • Start with read-only tools and a narrow allowlist. Add write or destructive tools only when you need them.
  • Assume a selected tool may receive the relevant prompt, file content, or conversation context needed to complete its call.
  • Review proposed sends, changes, deployments, purchases, deletions, and other consequential actions before allowing them.
  • Rotate service tokens on a schedule and immediately after suspected exposure.
  • Disable or delete a server that behaves unexpectedly; removing it prevents Chat from using its tools.

All available skill and MCP packs

The catalog currently includes 33 packs. Open Skills Marketplace for live versions, tool counts, installation sources, and new additions.

Official and verified MCP connections

Engineering, cloud, and operations

Productivity, data, research, and content

Travel

Community

Troubleshooting

ProblemCheck
A skill does not activateConfirm the pack and individual skill are installed and enabled, then describe a task that matches the skill's purpose.
An MCP server will not saveUse a valid https:// endpoint, a unique server label, and confirm the trust checkbox.
The connection saves but tools failConfirm the server supports Streamable HTTP or HTTP/SSE, the token is valid, and the allowed-tools names exactly match the server.
A tool is missingEdit the server's allowed tools, or leave the allowlist empty only when all of its tools are acceptable.
A token changedEdit the connection and enter the replacement token. Disable the connection until the new credential works.