Trace viewer

Debug how your agent reads every tool

Mockd captures the full conversation between your agent and its tools. Inspect each call, replay the session, and find exactly where the model misreads a tool — before it ever touches production.

sessionses_h7Qf…3a1

mock: HubSpot · v2024-10

crm.notes.create

429 errorinjected: 429 burst

started 12:04:02.005 · took 31 ms · 2 fields

Creates a note and attaches it to one or more CRM records.

From the tool catalog — what the agent read.

Agent inputrequired 1/1 · 0 unknown

I'm calling crm.notes.create with 2 values:

bodystringrequired

Text content of the note.

"Logged Q3 expansion deal from agent."

associationsarrayoptional

IDs of the records the note is attached to.

deal_9981

Exactly the data the agent sent (JSON)

{
  "body": "Logged Q3 expansion deal from agent.",
  "associations": ["deal_9981"]
}

Server response

what the tool returned

429 error

Rate limited — injected on purpose by the Mockd scenario „429 burst". Your agent waited 250 ms and the retry came back with 201 — from the mock, not the real tool.

{
  "error": "RATE_LIMITED",
  "message": "Too many requests. Injected by Mockd scenario 'burst'.",
  "retryAfterMs": 250
}

crm.contacts.searchquery: "acme"200142ms
crm.contacts.search
ok · 200
started 12:04:01.221 · took 142 ms · 3 fields
Full-text search across CRM contacts; returns matching records.
From the tool catalog — what the agent read.
Agent inputrequired 1/1 · 0 unknown
I'm calling crm.contacts.search with 3 values:
querystringrequired
Search term — e.g. a company name or email address.
"acme"
propertiesarrayoptional
Which contact fields to include in the result.
emailcompanylifecyclestage
limitnumberoptional
Maximum number of matches.
10
Exactly the data the agent sent (JSON)
{ "query": "acme", "properties": ["email", "company", "lifecyclestage"], "limit": 10 }
Server response
what the tool returned
200 ok
{ "total": 2, "results": [ { "id": "5512", "company": "Acme Inc.", "email": "ops@acme.io" }, { "id": "5513", "company": "Acme Labs", "email": "hi@acmelabs.dev" } ] }
crm.contacts.getid: 551220088ms
crm.contacts.get
ok · 200
started 12:04:01.402 · took 88 ms · 2 fields
Loads a single contact by its unique ID.
From the tool catalog — what the agent read.
Agent inputrequired 1/1 · 0 unknown
I'm calling crm.contacts.get with 2 values:
contactIdstringrequired
Unique ID of the contact.
"5512"
propertiesarrayoptional
Which fields to load.
emailcompanyownerlifecyclestage
Exactly the data the agent sent (JSON)
{ "contactId": "5512", "properties": ["email", "company", "owner", "lifecyclestage"] }
Server response
what the tool returned
200 ok
{ "id": "5512", "company": "Acme Inc.", "email": "ops@acme.io", "owner": "rep_42", "lifecyclestage": "opportunity" }
crm.deals.createname: "Acme – Q3 Expansion"201196ms
crm.deals.create
ok · 201
started 12:04:01.690 · took 196 ms · 4 fields
Creates a new deal in the CRM pipeline.
From the tool catalog — what the agent read.
Agent inputrequired 1/1 · 0 unknown
I'm calling crm.deals.create with 4 values:
dealnamestringrequired
Display name of the deal.
"Acme – Q3 Expansion"
amountnumberoptional
Deal value in the account's currency.
48000
pipelinestringoptional
ID of the pipeline the deal sits in.
"default"
dealstagestringoptional
Current stage in the sales process.
"qualifiedtobuy"
Exactly the data the agent sent (JSON)
{ "dealname": "Acme – Q3 Expansion", "amount": 48000, "pipeline": "default", "dealstage": "qualifiedtobuy" }
Server response
what the tool returned
201 ok
{ "id": "deal_9981", "dealname": "Acme – Q3 Expansion", "amount": 48000, "createdAt": "2026-06-19T12:04:01Z" }
crm.associations.createdeal_9981 ↔ 551220074ms
crm.associations.create
ok · 200
started 12:04:01.871 · took 74 ms · 3 fields
Links two CRM records together, e.g. a deal with a contact.
From the tool catalog — what the agent read.
Agent inputrequired 3/3 · 0 unknown
I'm calling crm.associations.create with 3 values:
fromobjectrequired
Record the association starts from — type (e.g. "deal") and id.
{object with 2 fieldstype:"deal"id:"deal_9981"}
toobjectrequired
Record being linked to — also type and id.
{object with 2 fieldstype:"contact"id:"5512"}
associationTypestringrequired
Type of link (e.g. deal_to_contact).
"deal_to_contact"
Exactly the data the agent sent (JSON)
{ "from": { "type": "deal", "id": "deal_9981" }, "to": { "type": "contact", "id": "5512" }, "associationType": "deal_to_contact" }
Server response
what the tool returned
200 ok
{ "status": "ASSOCIATED", "from": "deal_9981", "to": "5512" }
crm.notes.createinjectedrate limit — injected42931ms
crm.notes.create
429 errorinjected: 429 burst
started 12:04:02.005 · took 31 ms · 2 fields
Creates a note and attaches it to one or more CRM records.
From the tool catalog — what the agent read.
Agent inputrequired 1/1 · 0 unknown
I'm calling crm.notes.create with 2 values:
bodystringrequired
Text content of the note.
"Logged Q3 expansion deal from agent."
associationsarrayoptional
IDs of the records the note is attached to.
deal_9981
Exactly the data the agent sent (JSON)
{ "body": "Logged Q3 expansion deal from agent.", "associations": ["deal_9981"] }
Server response
what the tool returned
429 error
Rate limited — injected on purpose by the Mockd scenario „429 burst". Your agent waited 250 ms and the retry came back with 201 — from the mock, not the real tool.
{ "error": "RATE_LIMITED", "message": "Too many requests. Injected by Mockd scenario 'burst'.", "retryAfterMs": 250 }
crm.notes.create↻ retry okretried after 250ms backoff201110ms
crm.notes.create
ok · 201
started 12:04:02.379 · took 110 ms · 2 fields
Creates a note and attaches it to one or more CRM records.
From the tool catalog — what the agent read.
Agent inputrequired 1/1 · 0 unknown
I'm calling crm.notes.create with 2 values:
bodystringrequired
Text content of the note.
"Logged Q3 expansion deal from agent."
associationsarrayoptional
IDs of the records the note is attached to.
deal_9981
Exactly the data the agent sent (JSON)
{ "body": "Logged Q3 expansion deal from agent.", "associations": ["deal_9981"] }
Server response
what the tool returned
201 ok
{ "id": "note_4471", "createdAt": "2026-06-19T12:04:02Z" }

6 calls1 injectedtotal 641ms

Anatomy of a call

Everything captured, for every call

Each tool call your agent makes is recorded in full — so you see not just what happened, but why.

crm.deals.create

dealname: "Acme – Q3"
amount: 48000
pipeline: "default"

Tool & arguments

The exact tool the model picked and the arguments it filled in.

response

id: "deal_9981"
amount: 48000
createdAt: "…01Z"

Response

The synthetic payload Mockd returned — shaped like the real thing.

201 created196ms

0msbudget 300ms

Latency & status

How long the call took, and whether it succeeded, failed or was throttled.

forcedscenario: burst

429RATE_LIMITED

Injected scenario

Whether the response was forced by a scenario — a 429, a timeout, an empty set.

How it works

From config to first trace in under five minutes

No SDK to install, no code to change. Mockd looks exactly like a real MCP server to your agent — it just answers with synthetic data and records everything.

011 / 4

Pick a mock

Choose a server from the catalog — start with HubSpot — or bring your own MCP. You get a ready-to-paste connection config.

mock catalog8 servers

HubSpot24 tools
Salesforce31 toolssoon
Slack18 toolssoon
Stripe27 toolssoon
Notion15 toolssoon

022 / 4

Paste it into your agent

Drop the config into Claude, Cursor, the OpenAI Agents SDK or any MCP client. No code changes, no SDK to install.

mcp.jsonjson

{
  "mcpServers": {
    "mockd": {
      "url": "https://mock-d.com/hubspot/mcp"
    }
  }
}

033 / 4

Run your agent

Let it work. Every tool call is routed to the Mockd endpoint and answered with realistic, deterministic data.

agentmockd /mcpstreaming

crm.contacts.search200142ms
crm.contacts.get20088ms
crm.deals.create201196ms
crm.associations.create20074ms

044 / 4

Watch, then tune

Inspect each call in the live trace viewer, spot where the model misreads a tool, and fix your system prompt until it's right.

crm.deals.createmisread

model filled "name" — tool expects "dealname"

system prompt

− be a helpful CRM agent

+ use dealname (not name) when creating deals

✓ re-run · 201 created

Under the hood

A control room for your agent's tools

The pieces that turn 'it seemed to work' into a tight, observable feedback loop.

Full-fidelity trace viewer

Every tool call captured with arguments, response, latency and status — a LangSmith-style timeline built for MCP.

crm.contacts.search142ms

crm.deals.create196ms

crm.notes.create429

Error & latency injection

Flip a switch to return 429s, empty sets or slow responses and prove your agent recovers.

inject errors

429timeoutempty

Latency you control

Simulate real-world response times to see how your agent paces itself.

latency+250ms

Prompt diffing

Compare two system prompts across the same tool calls, side by side.

−be helpful

+prefer contactId over name

Stays in your sandbox

Nothing leaves to a third party. Mocked data is synthetic by design.

synthetic data

no egress

Bring your own MCP

Point Mockd at any custom MCP server. We render its tool descriptions, capture every call, and let you tune prompts against your own tools.

your-server.com/mcpconnected

tickets.listorders.getusers.search· 12 tools

See your agent's first trace

Point your agent at a Mockd endpoint and watch every tool call land in real time.

Start testing — free

sessionlive

crm.contacts.search142ms

crm.deals.create196ms

crm.notes.create110ms