# Easy Can Do MCP

Easy Can Do is a checked catalog of 1,264 AI models, APIs, data feeds and apps. Its MCP server lets a coding agent search the catalog while it builds, instead of searching the web.

- Server URL: https://easycando.com/mcp
- Transport: remote Streamable HTTP. Read-only.
- Account: none. No key and no sign-in.
- Limit: 300 requests an hour from one address. See Limits and errors.

## Set up

### Claude Code

Run this once in a terminal. It adds Easy Can Do to every project.

```
claude mcp add easycando --scope user --transport http https://easycando.com/mcp
```

Then start Claude Code and run `/mcp` to check that it’s connected.

### Cursor

Add this to `.cursor/mcp.json` in your home folder for every project, or in a project’s folder for just that project:

```
{
  "mcpServers": {
    "easycando": { "url": "https://easycando.com/mcp" }
  }
}
```

### Codex

```
codex mcp add easycando --url https://easycando.com/mcp
```

### Other clients

Add a remote HTTP MCP server named easycando with the URL https://easycando.com/mcp.

## Tools

- `search_tools`: finds tools for one step of a build. Give a short query, and have and want when you know what goes in and what should come out. Returns up to 10 previews, each with the capability that matched, its API status, price tier, the date it was checked and page link.
- `get_tool`: the full record of up to 5 tools. It lists each capability with its price and limits, access and auth, ways in, docs and pricing links, terms, and the date it was checked.
- `find_chains`: finds tools that turn one kind of input into another, alone or two in a row.

Each reply holds a text block for the agent to read, and the same results as JSON in structuredContent.

## How an agent uses it

1. Split the build into steps. Call search_tools once per step.
2. Call get_tool on the picks before writing integration code.
3. When no single tool does a step, call find_chains.
4. Tell the user which tool it picked and why, and link the tool’s page.

## Reference

The sections below say exactly what each tool accepts and returns. The example replies were captured from a local copy of the server on 2026-11-18, with the full catalog. The tools and their order will change as the catalog and the ranking change.

### Calling the server

Send each call as an HTTP POST to https://easycando.com/mcp with a JSON-RPC body and these two headers:

```
Content-Type: application/json
Accept: application/json, text/event-stream
```

The server keeps no session, so a raw call needs no initialize first. A client library does the handshake for you. A request that does not accept both types gets 406.

The reply is a text/event-stream with one message event. Its data line is a JSON-RPC result that holds a text block and structuredContent. Both carry the same results.

- The text block is written for an agent to read. Its layout can change.
- structuredContent is JSON for code. It follows the outputSchema that tools/list gives for each tool. Each tool’s section below lists its fields.
- A client may show its model either one.

A call with bad input still gets HTTP 200. The result then has `"isError": true` and text that says what to fix. An error has no structuredContent.

### have and want

have and want take one of these ids. Each id groups several of the type ids that tool pages show for a capability, as in `(audio-speech, url → text, structured)`.

| Id | What it covers | Type ids it groups |
| --- | --- | --- |
| words | Words | text, query, message, social-post |
| photos | Photos and images | image, image-mask, 3d |
| video | Video | video |
| sound | Voice and sound | audio-speech, audio, audio-music, audio-sfx |
| documents | Documents | document |
| links | Web links | url, webpage |
| places | Places | geo |
| data | Numbers and data | structured, timeseries, event-stream, embedding |
| code | Code | code |
| money | Money | money |
| people | People | profile |
| dates | Dates and events | schedule |
| done | Gets something done. Only for want. | action:publish, action:message, action:money, action:world, action:software, action:browser, action:schedule, action:record, physical-item |

- have and want also take a type id from the third column. A type id stands for the id in its row. have audio-speech is the same call as have sound, so it also finds tools that take music or sound effects.
- The next-page arguments name the id. After have audio-speech, they say have sound.
- have takes any id or type id except done and the type ids in its row. want takes any of them. done means the tool acts: it posts, sends, pays, books or changes something.
- Any other value fails with an input validation error. The error lists every value that side takes, as the last example in this section shows.
- A tool matches when one capability takes the have id and gives the want id. Both must hold in the same capability.
- Numbers and data counts as something a tool gives only when it is the first output listed for the capability. Most tools list structured data after their real output.
- In search_tools, leave have or want out to match anything on that side. find_chains needs both.

A call with a value that is not an id or a type id:

```
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_tools","arguments":{"query":"transcribe audio","have":"speech"}}}
```

The reply:

```
Input validation error: Invalid arguments for tool search_tools: have: Invalid option: expected one of "words"|"photos"|"video"|"sound"|"documents"|"links"|"places"|"data"|"code"|"money"|"people"|"dates"|"text"|"query"|"message"|"social-post"|"image"|"image-mask"|"3d"|"audio-speech"|"audio"|"audio-music"|"audio-sfx"|"document"|"url"|"webpage"|"geo"|"structured"|"timeseries"|"event-stream"|"embedding"|"profile"|"schedule"
```

### search_tools

Give a query, have, want, or any mix. A call with none of them is an error.

| Name | Type | Default | Meaning |
| --- | --- | --- | --- |
| query | string, up to 200 characters | none | What the step does, in a few words. |
| have | one of: words, photos, video, sound, documents, links, places, data, code, money, people, dates, or a type id they group | none | What the step starts with. |
| want | an id or type id from the list above | none | What the step should produce. |
| free | boolean | no filter | Only tools that are free or have a free tier. |
| local | boolean | no filter | Only tools that run on the user’s own machine. |
| live | boolean | no filter | Only tools with live or real-time data. |
| official | boolean | no filter | Only tools with an official API. |
| group | one of: images, video, voice, design, ai, reading, web, social, knowledge, world, posting, money, home, mac, building | none | Only one group. |
| limit | integer, 1 to 10 | 5 | Results per page. |
| offset | integer, 0 to 1,000 | 0 | Results to skip. |

Each result is three lines:

1. The number, the tool’s name, its slug in parentheses, then its one-line summary.
2. Does: the capability that matched, then its type ids as (inputs → outputs).
3. The API status, then the price tier (free, freemium, low, mid, high or enterprise; see Price tiers), then runs locally and live data when they apply, then the date it was checked, then the page address.

A reply with more results has a More: line. It holds the arguments for the next page. Send them as they are. No More: line offers an offset over 1,000.

search_tools returns strong matches only. A looser match covers only part of the query, or matches only through a side feature. The reply leaves looser matches out of the results and the total. A line such as `24 looser matches are left out` gives their count.

structuredContent holds the same page:

| Field | Meaning |
| --- | --- |
| total | Results in all, across every page, looser matches left out. With fallback true, the tools the filters leave. |
| offset | Results skipped before this page. |
| results | The tools on this page, best first, with the result fields below. |
| looser | Looser matches left out. 0 when there are none. |
| fallback | true when nothing matched the query, and results are the most notable tools the filters leave. |
| note | The note at the top of the text, such as `Nothing matched …`, or null. |
| next | The arguments for the next page, or null on the last page. Send them as they are. |

Result fields. search_tools results, find_chains one_step and each step of a chain use them:

| Field | Meaning |
| --- | --- |
| slug | The tool’s slug. See Slugs. |
| name | The tool’s name. |
| sentence | Its one-line summary. |
| capability | The capability that matched, as index, line, in and out. When the query matched only the tool’s name or summary, it is the first capability that passes have and want, else the tool’s first. index is its position in the tool’s capabilities in get_tool, from 0. in and out are type ids. null when a tool lists no capability. |
| api_status | official-ga, official-beta, waitlist, partner-or-enterprise, unofficial or none. |
| price_tier | free, freemium, low, mid, high or enterprise. See Price tiers. |
| local | true when the tool runs on the user’s own machine. |
| live | true when it has live or real-time data. |
| checked_on | The date the tool was last checked, as YYYY-MM-DD. |
| page_url | The tool’s easycando.com page. |

Example request, as one raw exchange:

```
curl -s https://easycando.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_tools","arguments":{"query":"transcribe audio","have":"sound","want":"words","limit":2}}}'
```

The raw reply (the text field holds the line breaks as \n):

```
event: message
data: {"result":{"content":[{"type":"text","text":"25 tools match. Showing 1 to 2.\n\n1. ElevenLabs Scribe (elevenlabs-scribe): Transcribe audio with speakers labeled and laughter or applause tagged.\n   Does: transcribe recorded audio with diarization, audio-event tags and entities (audio-speech, url, video → text, structured).\n   Official API, price low, checked 2026-11-03. https://easycando.com/tools/elevenlabs-scribe\n2. Deepgram (deepgram): Transcribe speech live and know when the speaker has finished talking.\n   Does: transcribe recorded audio with timestamps and diarization (audio-speech, url → text, structured).\n   Official API, price low, checked 2026-11-03. https://easycando.com/tools/deepgram\n\nMore: call search_tools with {\"query\":\"transcribe audio\",\"have\":\"sound\",\"want\":\"words\",\"limit\":2,\"offset\":2}\n24 looser matches are left out: they match only part of the query, or only a side feature. Try other words to see more.\nThese are previews. Call get_tool on the tools you pick before writing integration code."}],"structuredContent":{"total":25,"offset":0,"results":[{"slug":"elevenlabs-scribe","name":"ElevenLabs Scribe","sentence":"Transcribe audio with speakers labeled and laughter or applause tagged.","capability":{"index":0,"line":"transcribe recorded audio with diarization, audio-event tags and entities","in":["audio-speech","url","video"],"out":["text","structured"]},"api_status":"official-ga","price_tier":"low","local":false,"live":false,"checked_on":"2026-11-03","page_url":"https://easycando.com/tools/elevenlabs-scribe"},{"slug":"deepgram","name":"Deepgram","sentence":"Transcribe speech live and know when the speaker has finished talking.","capability":{"index":0,"line":"transcribe recorded audio with timestamps and diarization","in":["audio-speech","url"],"out":["text","structured"]},"api_status":"official-ga","price_tier":"low","local":false,"live":false,"checked_on":"2026-11-03","page_url":"https://easycando.com/tools/deepgram"}],"looser":24,"fallback":false,"note":null,"next":{"query":"transcribe audio","have":"sound","want":"words","limit":2,"offset":2}}},"jsonrpc":"2.0","id":1}
```

The same text, with its line breaks:

```
25 tools match. Showing 1 to 2.

1. ElevenLabs Scribe (elevenlabs-scribe): Transcribe audio with speakers labeled and laughter or applause tagged.
   Does: transcribe recorded audio with diarization, audio-event tags and entities (audio-speech, url, video → text, structured).
   Official API, price low, checked 2026-11-03. https://easycando.com/tools/elevenlabs-scribe
2. Deepgram (deepgram): Transcribe speech live and know when the speaker has finished talking.
   Does: transcribe recorded audio with timestamps and diarization (audio-speech, url → text, structured).
   Official API, price low, checked 2026-11-03. https://easycando.com/tools/deepgram

More: call search_tools with {"query":"transcribe audio","have":"sound","want":"words","limit":2,"offset":2}
24 looser matches are left out: they match only part of the query, or only a side feature. Try other words to see more.
These are previews. Call get_tool on the tools you pick before writing integration code.
```

The same structuredContent, with its line breaks:

```
{
  "total": 25,
  "offset": 0,
  "results": [
    {
      "slug": "elevenlabs-scribe",
      "name": "ElevenLabs Scribe",
      "sentence": "Transcribe audio with speakers labeled and laughter or applause tagged.",
      "capability": {
        "index": 0,
        "line": "transcribe recorded audio with diarization, audio-event tags and entities",
        "in": [
          "audio-speech",
          "url",
          "video"
        ],
        "out": [
          "text",
          "structured"
        ]
      },
      "api_status": "official-ga",
      "price_tier": "low",
      "local": false,
      "live": false,
      "checked_on": "2026-11-03",
      "page_url": "https://easycando.com/tools/elevenlabs-scribe"
    },
    {
      "slug": "deepgram",
      "name": "Deepgram",
      "sentence": "Transcribe speech live and know when the speaker has finished talking.",
      "capability": {
        "index": 0,
        "line": "transcribe recorded audio with timestamps and diarization",
        "in": [
          "audio-speech",
          "url"
        ],
        "out": [
          "text",
          "structured"
        ]
      },
      "api_status": "official-ga",
      "price_tier": "low",
      "local": false,
      "live": false,
      "checked_on": "2026-11-03",
      "page_url": "https://easycando.com/tools/deepgram"
    }
  ],
  "looser": 24,
  "fallback": false,
  "note": null,
  "next": {
    "query": "transcribe audio",
    "have": "sound",
    "want": "words",
    "limit": 2,
    "offset": 2
  }
}
```

### get_tool

| Name | Type | Default | Meaning |
| --- | --- | --- | --- |
| slugs | list of 1 to 5 strings, each up to 300 characters | required | The tools to return. |

Each string is a slug, an old slug (see Slugs), an easycando.com page address with or without .md, or an exact tool name. The same tool named twice comes back once.

The reply holds one record per tool, split by a line of three hyphens. A record is the same text as the tool’s .md page. A string that matches no tool gets a line like `No tool “deepgrm”. Did you mean: deepgram?`. When none match, the result has isError set. A record longer than 6,000 characters stops its list of capabilities early, with a last line like `(3 more on <page address>)`.

structuredContent has two fields:

- tools: one record per tool, with the same facts as the text record. Values are raw ids, such as api-key for auth. A field with no value is null.
- missing: each string that matched no tool, as ref, with up to three close slugs as suggestions.

A record has these fields: slug, name, sentence, page_url, provider, group, kind, api_status, signup, auth, price_tier, free_tier, ways_in, also_through, docs_url, pricing_url, site_url, local, live, update_frequency, terms, ranking, ranking_source, no_safe_way, safer_routes, checked_on and capabilities.

- Each capability has index (from 0), line, in, out, mode, price and limits. The JSON lists every capability, even when the text stops early.
- safer_routes lists the ways in for a tool with no official API, each with its risk. no_safe_way is true when the catalog found no safe way in.

Example request:

```
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_tool","arguments":{"slugs":["deepgram"]}}}
```

The reply text, in full:

```
# Deepgram

Transcribe speech live and know when the speaker has finished talking.

- Page: https://easycando.com/tools/deepgram
- Provider: Deepgram. Group: Voice and music. Kind: API.
- Access: Official API. Sign-up: Sign up yourself. Auth: API key.
- Price tier: low. Free tier: $200 non-expiring credit, no card (~43k Nova-3 minutes)
- Ways in: web API, live stream, Python, JavaScript.
- Docs: https://developers.deepgram.com/docs/models-languages-overview · Pricing: https://deepgram.com/pricing · Site: https://deepgram.com
- Runs locally: no. Live data: no.
- Terms: Commercial use allowed.
- Ranking: AA STT speed-factor #2 (Nova-3, 564x realtime), Artificial Analysis, checked 2026-11-03
- Checked on 2026-11-03. Prices and limits change; check the provider’s pages before you rely on them.

## What it can do
1. transcribe recorded audio with timestamps and diarization (audio-speech, url → text, structured). Mode: sync. Price: $0.0043/min (Nova-3 mono, prerecorded); $0.0052/min multilingual. Limits: per-second billing, no minimums.
2. stream live transcription with end-of-turn detection for voice agents (audio-speech → text, event-stream). Mode: streaming. Price: $0.0077/min (Flux English or Nova-3 streaming). Limits: WebSocket.
3. summarize, detect topics, sentiment and intent from transcripts (audio-speech, text → structured, text). Mode: sync. Price: $0.0003/1k input tokens + $0.0006/1k output tokens on top of STT.
```

### find_chains

| Name | Type | Default | Meaning |
| --- | --- | --- | --- |
| have | one of: words, photos, video, sound, documents, links, places, data, code, money, people, dates, or a type id they group | required | What the chain starts with. |
| want | an id or type id from the list above | required | What the chain should produce. |
| query | string, up to 200 characters | none | What the chain is for. Results rank by it. Without it they rank by notability alone, and the reply says so. |
| limit | integer, 1 to 10 | 5 | How many single tools, and how many two-tool chains, to list. |

There is no offset. A chain has at most two tools.

The reply has two lists. The first holds the tools that do the whole step in one capability, in the three-line form above. The second holds two-tool chains. Each shows the first tool, the type it hands over, and the second tool, then what each one does. The chain lines carry slugs but no checked date, so call get_tool on the tools you pick.

structuredContent holds the same lists:

| Field | Meaning |
| --- | --- |
| have, want | The call’s have and want, as ids. A type id comes back as the id in its row. |
| ranked_by | query, or notability when the call has no query or its query has no words to rank by. |
| one_step_total | Tools that do it in one capability, in all. |
| one_step | The first of them, best first, with the result fields. |
| two_steps_total | Two-tool chains, in all. |
| two_steps | The first chains, best first. Each has first, middle and second. first and second have the result fields, checked_on included. middle is the type id the first tool hands to the second. |

Example request:

```
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"find_chains","arguments":{"have":"video","want":"done","query":"post clips to tiktok","limit":2}}}
```

The reply text:

```
From video to a done action.

One tool does it (34 in all, best first):
1. Submagic (submagic): Add animated captions and hook titles, then post to six networks.
   Does: publish or schedule an exported project to YouTube, TikTok, Instagram, X, Facebook, LinkedIn; read post analytics (video, text → action:publish, structured).
   Official API, price low, checked 2026-11-03. https://easycando.com/tools/submagic
2. OpusClip (opusclip): Turn a long video into ranked viral shorts and post them.
   Does: generate social copy and publish or schedule clips to connected accounts (video, text → action:publish).
   Official API, price mid, checked 2026-11-03. https://easycando.com/tools/opusclip

Two tools in a row (8003 chains in all, best first):
1. Gemini Omni (gemini-omni) → video → Submagic (submagic)
   First: extend a clip at its end, including continuing spoken dialogue. Then: publish or schedule an exported project to YouTube, TikTok, Instagram, X, Facebook, LinkedIn; read post analytics.
2. Hailuo (hailuo) → video → Submagic (submagic)
   First: regenerate a 768P clip at 2K. Then: publish or schedule an exported project to YouTube, TikTok, Instagram, X, Facebook, LinkedIn; read post analytics.

Call get_tool on the tools you pick before writing integration code.
```

### Slugs

A slug is a tool’s name in this server: lowercase letters, digits and hyphens, as in deepgram. Every search result, chain and record shows it, and the tool’s page address ends with it.

- A slug can change when a tool is renamed. The old slug keeps working. get_tool returns the current record for it, and the old page address, with or without .md, redirects with a 308 to the new one.
- To learn a tool’s current slug, read the Page line of its record. Store that one.
- A slug is never given to a different tool. An old slug and a retired tool’s slug stay taken, and the data check that runs before each build fails if a tool reuses one.
- A retired tool is gone from search_tools, find_chains and get_tool, which answers No tool “slug”. Its page stays up with a notice. Its .md copy returns 404.
- Capabilities have no ids. The numbers in a text record and index in structuredContent are list positions. The text counts from 1, and index counts from 0. A position can change when the catalog changes.

### Price tiers

Each tool has one price tier. The tier sorts the provider’s prices into a few bands, for filtering. Where a provider publishes no prices, the tier is an estimate.

| Tier | What it costs | Tool pages show |
| --- | --- | --- |
| free | No charge for the main way in. Some also sell a paid hosted plan. | free |
| freemium | A free tier, then paid use. free_tier in get_tool says what the free tier gives. | free tier |
| low | A hobby budget: under about $20 a month, or cents per call. | $ |
| mid | About $20 to $200 a month. | $$ |
| high | Over about $200 a month, or dollars per call. | $$$ |
| enterprise | Priced for organizations, mostly through sales or a partner program. Most publish no price. | enterprise |

A tier is a filter, not a quote. The capabilities in get_tool give the provider’s own prices. Check them on the provider’s pages before you rely on them.

### Limits and errors

- 300 requests an hour from one address. Every POST counts: tool calls, initialize, tools/list and notifications. A batch counts each message in it, and a batch holds at most 10.
- The hour is the clock hour in UTC. The count restarts at the top of the hour. It is not a sliding window.
- An address is the caller’s IP address. IPv6 addresses share a limit by their first 64 bits. A CI farm or an office network behind one address shares one limit.
- 100,000 requests a day across all addresses, counted by UTC date. A request that the hourly limit blocks does not count toward it.

Over the hourly limit, the server answers with HTTP 429 Too Many Requests and a Retry-After header. Retry-After is the number of seconds to the top of the next hour. This is the exchange for the 301st request in an hour:

```
HTTP/1.1 429 Too Many Requests
content-type: application/json
retry-after: 2115

{"jsonrpc":"2.0","error":{"code":-32000,"message":"That’s more than 300 calls this hour from one address. Try again in 36 minutes."},"id":null}
```

Over the daily cap, the server answers with HTTP 503 Service Unavailable and Retry-After set to 3600. The cap restarts at 00:00 UTC, so a retry after an hour may get the same answer:

```
HTTP/1.1 503 Service Unavailable
content-type: application/json
retry-after: 3600

{"jsonrpc":"2.0","error":{"code":-32000,"message":"Easy Can Do is busy today. Try again tomorrow."},"id":null}
```

These are HTTP errors with a JSON-RPC error body and no result. Wait for the Retry-After seconds, then send the same request again. The limits can change. The Retry-After header is the current wait.

Other errors:

| Status | When | Body message |
| --- | --- | --- |
| 200, with isError | The arguments are wrong, or a tool reference matches nothing. | Text that says what to fix. |
| 400 | A batch holds more than 10 calls. | Send at most 10 calls in one batch. |
| 406 | Accept does not list both application/json and text/event-stream. | Not Acceptable: Client must accept both application/json and text/event-stream |
| 413 | The request body is over 16,384 bytes. | That request is too large. |

## Without the MCP

Every tool page has a markdown copy: add .md to its address, as in https://easycando.com/tools/deepgram.md. The index for agents is https://easycando.com/llms.txt.

## Privacy

The server keeps no query text. It logs the tool name, the number of results and the time taken. The hourly limit counts calls under a salted hash of the address, which expires within two hours.
