MCP servers

​

Attaching

{
  "config": {
    "mcps": [{ "integration_id": "7c2f1a90-4e83-4d16-9b7a-2c5e08f31d64" }]
  }
}

AGENT_ID=0f9a1c3e-5d42-4b8a-9c11-6e2f7a4d8b30
INTEGRATION_ID=7c2f1a90-4e83-4d16-9b7a-2c5e08f31d64

curl -s "https://api.talqing.com/v1/agents/$AGENT_ID" \
  -H "Authorization: Bearer $TALQING_API_KEY" \
| jq --arg id "$INTEGRATION_ID" \
    '{config: {mcps: (.config.mcps + [{integration_id: $id}])}}' \
| curl -s -X PATCH "https://api.talqing.com/v1/agents/$AGENT_ID" \
    -H "Authorization: Bearer $TALQING_API_KEY" \
    -H "Content-Type: application/json" \
    --data @-

curl -X POST "https://api.talqing.com/v1/agents/$AGENT_ID/publish" \
  -H "Authorization: Bearer $TALQING_API_KEY"

const agent_id = "0f9a1c3e-5d42-4b8a-9c11-6e2f7a4d8b30";
const integration_id = "7c2f1a90-4e83-4d16-9b7a-2c5e08f31d64";

const { config } = await talqing.agents.get({ agent_id });

await talqing.agents.update({
  agent_id,
  updateAgentRequest: { config: { mcps: [...config.mcps, { integration_id }] } },
});
await talqing.agents.publish({ agent_id });

agent_id = "0f9a1c3e-5d42-4b8a-9c11-6e2f7a4d8b30"
integration_id = "7c2f1a90-4e83-4d16-9b7a-2c5e08f31d64"

mcps = talqing.agents.get(agent_id)["config"]["mcps"]

talqing.agents.update(agent_id, config={"mcps": [*mcps, {"integration_id": integration_id}]})
talqing.agents.publish(agent_id)

​

Listing a server’s tools

curl "https://api.talqing.com/v1/integrations/$INTEGRATION_ID/mcp-tools" \
  -H "Authorization: Bearer $TALQING_API_KEY"

{
  "tools": [
    { "name": "search", "exposed_name": "crm_search", "description": "Find a contact by name, email or phone number." },
    { "name": "create_note", "exposed_name": "crm_create_note", "description": "Attach a note to a contact." }
  ]
}

• name is the server’s own name. allowed_tools matches this.
• exposed_name is what the model sees. Your prompt must use this.

​

Approving tools

• null (the default for a new connection) exposes every tool.
• A PATCH replaces the list wholesale. To go back to everything, send the full list from mcp-tools; there is no reset value.
• It cannot be empty. To approve nothing, disable the integration.
• Entries are the tools’ own names, not exposed_names, so changing the namespace never invalidates an approval.

curl -X PATCH "https://api.talqing.com/v1/integrations/$INTEGRATION_ID" \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"allowed_tools": ["search", "create_note"]}'

​

Namespacing

• Default: the provider key for hosted providers (asana, google_calendar), "" for custom_mcp.
• "" means “use the server’s own names”. On a PATCH, omitting the field keeps the current value and "" clears it.
• A tool whose name already starts with the prefix is not prefixed again: Tavily’s tavily_search stays tavily_search.
• Pattern: ^$|^[a-z][a-z0-9_]{0,23}$.

'Asana (2)' and 'Asana' both present their tools under 'asana' - every tool
name they share would reach the model twice. Give one of them a different tool
namespace.

​

What happens on a call

• Servers connect when the session starts, in parallel. The connect, the tool listing and each tool call have a five-second timeout.
• A server that fails to connect does not fail the call. The agent starts without its tools, and the model will improvise around a prompt that mentions them. Tell the prompt what to do when a lookup is unavailable.
• A tool call that times out or errors returns to the model as a failed tool call.
• OAuth providers: a 401 triggers one token refresh and retry; a second failure moves the integration to needs_reconnect.
• Native Google Calendar tools call Google directly with their own retries.

​

Field reference

​

Next